essay in role model

Essays About Role Models: Top 5 Examples and 7 Prompts

Check our guide, including top examples and prompts on essays about role models to write a high-quality piece about your idol and motivator.

How do you label someone a “role model”? Do you look for exemplary qualities like high intelligence, good leadership skills, and confidence? Role models are critical in personality development and life. For example, research proves that adolescents with positive role models, primarily those they know personally, do better academically and have more courage and self-respect.

A role model is someone who excels in their field of expertise, inspires you to be a better person, and motivates you to follow an excellent path. They influence your life decisions and encourage you to become a lifelong learner. Who individuals decide to be their role models vary. Although most pick their parents , others also choose prominent people like celebrities, politicians, or singers.

If you want to start writing about your role model, here are five sample essays to inspire you:

1. Best Role Model in Life by Prasanna

2. my role model: essay about my father by anonymous on gradesfixer.com, 3. word essay on true role models by pragati ghosh, 4. the role model qualities of martin luther king junior: leadership, courage, and self-discipline by anonymous on studymoose.com, 5. my mother: my inspiration and my role model by ankita yadav, 1. why should i have a role model, 2. my role models, 3. what is a role model, 4. disney characters and role models, 5. celebrities and singers as role models, 6. characteristics of a good role model, 7. how do role models affect young people.

“But for me, a role model is someone who motivates me and influences me to be a good human and positive person in life, who gives me a supporting hand to walk and run to achieve my dreams and goals. Role model is not someone who is full of ego, attitude or full of selfishness.”

Prasanna introduces her role model A.R. Rahman as someone who stimulates her desire to be an excellent musician. She mentions how vital it is to look for a role model in the field that you’re interested in. For her, it’s music. Rahman stirs up her passion for learning guitar and reaching the same achievements he had.

She believes anyone can be someone’s role model as long as they are a good person. Prasanna adds that a good role model must help you reach your dreams. She also shares her wish to meet her role model to thank him personally.

Are you interested in music? Check out these 10 essays about music .

“He is the one I always admire in my life. When I have children, I want to be similar to my father to do for my kids how he takes care of me.”

The essay shows the excellent characteristic of the author’s role model, their father. They describe him as a knowledgeable person who always encourages them to be strong in every situation and supports their goals in life. Further, the writer mentions that their father earns respect and admiration from others because of his hard work in becoming a successful businessman and a reputable human being.

“A role model is somebody who has influenced us in our lives and whom we try to imitate and learn as we are aspired by them.”

Ghosh’s essay acknowledges that in today’s era, a person’s character, style, body language, virtues, deeds, and actions influence what type of role model they will be for others. She notes that an actual role model has self-confidence and good communication skills, shares similar thoughts with those they inspire, knows how to get credit, and does things with good intentions. She concludes her essay by advising the readers to choose their role models wisely.

“… He was a civil right activist and is one of the reasons why African Americans have rights.”

This essay expounds on Martin Luther King Jr .’s qualities that the author admires. These characteristics pushed the writer to choose him as their most significant role model. The piece recounts how Luther King Jr.’s leadership skills in fighting for equal rights and courage in speaking in front of police officers inspired the writer to be as brave as the activist and Baptist minister. The author also includes that despite being busy, Luther King Jr. finished his work on time because of self-discipline.

“She is a mother, a pilot, a driver, a cook, a good person, all at the same time. I not only want to be a pilot like her but want to copy her overall personality.”

In Yadav’s essay, she discusses how her mother became her role model. Aside from being a hard worker, she points out how she religiously follows schedules to be timely with her firm self-discipline. She believes her mother isn’t only the next Master Chef but also someone with a high sense of fashion. She’s a busy person with style. Yadav’s role model is with her at home – a mother, a mentor, and a perfect friend with a fantastic personality.

7 Writing Prompts for Essays About Role Models

Are you ready to write an essay about your role model? If you don’t have a role model or you’re having trouble coming up with a topic to start your essay writing, feel free to use the prompts below:

Most people, especially students, look for a role model to motivate them to find the right career path. At the same time, this path must also let them focus on goals, develop skills, and build character. Use this prompt to discuss your reasons why you want to have or have a role model. Then, introduce your role model by sharing their characteristics and what makes them special. You might also be interested in these essays about critical thinking.

Essays about role models: My role models

People have more than one role model. For this prompt, list your heroes and why you consider them as such. Then, pick the top ones and add their lessons’ impact on your life. Finally, you can include some of their inspirational quotes and explain why these sayings resonate with you.

Define what role models are all about and link relevant research or statistics that prove their importance. Then, talk about your understanding of being a role model and what qualities most look for in their role models today.

Disney characters, especially princesses, are what young girls want to be. Share your stance on the debate of whether it’s okay for kids to use a fictional character as their role model or not. Affix your reasons why. Make your essay more convincing by adding examples, especially if you disagree with the idea that imaginary characters are worth being idolized.

Many look up to different celebrities and singers and consider them their role models. Some take it to the extreme and follow exactly what their idols do. Unfortunately, many become obsessed and follow even the wrong things, thinking they will gather the same fame and status as their idols. They also forget to live their own unique life. First, say something about youngsters worshipping popular artists. Then, add how people can select more positive and influential models to adore.

Discuss the typical characteristics people associate when they think of a “role model.” Then, give examples of these ideal individuals and share your preferences when favoring a role model.

Role models should motivate and inspire the youth to reach their goals and live as good people. Open discourse about the benefits of having a positive role model and how they can help change the unhealthy behaviors of today’s youth. Make your piece more compelling by adding the adverse effects of having a negative role model.

Excellent grammar and punctuation are crucial to ensuring a good grade in your essay. See our guide to grammar and punctuation to know more.

For help with your essay, check our round-up of best essay writing apps .

Maria Caballero is a freelance writer who has been writing since high school. She believes that to be a writer doesn't only refer to excellent syntax and semantics but also knowing how to weave words together to communicate to any reader effectively.

View all posts

Role Model Essay for Students and Children

500+ words essay on role model.

Role Model Essay – There are many people in the world who blindly follow anyone. Also, they admire and love that person no matter what he does. A role model is a person who inspires you to be like him.

Moreover, a role model is the person you love and want to be like him. Besides this, the role model can be anyone from a celebrity to a politician to your family member.

My Role Model

The missile man of India Dr. APJ Abdul Kalam is my role model. He is my role model not because he has achieved great things in life and become the president of India. He is my role model because he dedicated his entire life to the country and its people especially school children.

His Achievement

Dr. APJ Abdul Kalam worked his entire life and achieved many great things in life. But several of his bigger achievements are the testing of Atomic bombs at Pokhran, development of the missile Agni and Prithvi, and becoming the president of India . But there is one more thing which is of great importance that is after his retirement from the presidential post he started teaching because it was his passion.

He always says that this young generation is the future of the country so guide them on the right path and they lead India to greatness.

His Personal Life

Dr. Abdul Kalam was so dedicated to his work that he usually sleeps late at night and wakes up early in the morning. Also, he never misused his power as a president of India. Besides, all this his family (brother and sisters) was also a very humble person who earn their living by doing work and they live simple lives.

Get the huge list of more than 500 Essay Topics and Ideas

Qualities of a Role Model

There are certain qualities of a good role model which make him/her a great man/women. These qualities include a demonstration of confidence and leadership , communication with everyone, being unique, showing respect and concern for others, treating everyone equally, willingness to admit mistakes, and maintaining the same character inside or outside the job. If a person possesses these qualities then he/she can be an ideal role model.

How to be a Role Model for Others?

Being a role model to someone is not an easy task especially if you focus on being a role model. The thing about being a role model is that you only do the right thing and people starts to follow. Also, you do not have to achieve something bigger just follow your heart and the above-mentioned qualities and be patient and you will be a role model one day.

My Other Role Model

Apart from Dr. Abdul Kalam, I have other role models in my life that belong to different fields and different profession. The persons are Ratan Tata, Bill Gates, Will Smith, Sachin Tendulkar, Kapil Dev, Kalpana Chawla , Albert Einstein , B.R. Ambedkar, my father, and my mother.

{ “@context”: “https://schema.org”, “@type”: “FAQPage”, “mainEntity”: [{ “@type”: “Question”, “name”: “Can a person have more than one role model?”, “acceptedAnswer”: { “@type”: “Answer”, “text”: “Yes, it is not mandatory for anyone to have only one role model. You role model can be different persons, different field, and different professions.” } }, { “@type”: “Question”, “name”: “Whom we can call a good role model?”, “acceptedAnswer”: { “@type”: “Answer”, “text”:”The person who confident, and helping, who communicates with everyone, respect others, treat everyone equally, influence other, accept his/her mistakes, and maintain a balanced character with everyone is the ideal role model for everyone.”} }] }

Customize your course in 30 seconds

Which class are you in.

Travelling Essay
Picnic Essay
Our Country Essay
My Parents Essay
Essay on Favourite Personality
Essay on Memorable Day of My Life
Essay on Knowledge is Power
Essay on Gurpurab
Essay on My Favourite Season
Essay on Types of Sports

Download the App

Admission Essay
Statement of Purpose Editing
Personal Statement Editing
Recommendation Letter
Motivation Letter
Cover Letter
Supplemental Essay
Letter of Continued Interest
Scholarship Essay
Role Model Essay
Our Editors
College Admission Essay Examples
College Cover Letter Examples
College Personal Statement Examples
Graduate Personal Statement Examples
Graduate Statement of Purpose Examples
MBA Essay Examples
MBA Personal Statement Examples
MBA Resume Examples
MBA Recommendation Letter Examples
Medical School Personal Statement Examples
Medical School Recommendation Letter Examples
Pricing Plans
Public Health
Dissertation
Research Paper
Thesis Editing
Academic Editing
Motivation letter
Letter of Recommendation
Personal Statement
Statement of Purpose

My Role Model Essay: A Few Ways to Elaborate on The Subject

EssayEdge > Blog > My Role Model Essay: A Few Ways to Elaborate on The Subject

You might be surprised at how much your answer to this question can say about you. The role model you choose for your role model essay is not nearly as important as your description of that person and the way you relate this back to yourself. Here are some principles to keep in mind when writing this essay.

Choose someone with whom you have had a meaningful relationship. Who is your role model? Don’t pick a more important but distant figure. No one will be impressed if you choose, for example, the CEO of your company but have nothing to say about him beyond listing the credentials in his company profile. Note: There are some questions that allow you to identify a “hero,” and others that ask you to describe an ideal manager. In those cases you should alter your choice accordingly, but the below principles still apply.
Express thoughtful admiration. Glorifying your role model makes you sound naíve. Focus instead on describing the person in an insightful manner and explaining why you value what he or she represents. You might even have reason to differentiate yourself from your role model in specific ways, though the overall purpose should be to describe what you admire and want to emulate.
Describe and illustrate specific qualities. My role model essays should be accurate. Avoid relying on vague language that merely depicts your role model as “brilliant” or “compassionate.” Instead, aim for a more penetrating portrait that captures your role model’s complexities. Moreover, just because you’re not writing about yourself does not mean that you don’t have to back up your claims. Show your role model in detailed action to convey the qualities that he or she embodies.

Tell stories. Using specific anecdotes is the best way to achieve depth in your description. By describing the details of a particular episode, you can convey much more about a person than by listing any number of characteristics.
Show the role model’s tangible influence on you. The best way to demonstrate the importance your role model has played in your life is to recount changes you’ve made based on that person’s influence. Of course, you should not come across as a sycophant or a blind follower. But you can show how your role model provoked ideas that then initiated a course of action.

Need help? Check out EssayEdge editing services:

This applicant offers a meaningful portrait of his father in just 300 words. Unlike the previous applicant , he chooses to name the qualities that he admires right from the outset, but he then goes on to justify them with concrete examples.

How Role Models Influence Behaviors Essay

Introduction, addictive behavior, psychology of addictive behavior, aims of socialisation, process of socialisation.

A role is a behavior or attention to duty expected of an individual to which he or she may adhere to another. A role model is a character implementing a certain role in society for others to copy. This is to say a role model is used as an example by other actors in the society. They can be teachers in various institutions whose characters are admired by students acting as role models to their children or even vice versa.

Role models are perceived differently in each society and even each individual or age group, one’s role model may not be another’s or one society’s role model may differ from another. A certain age group for instance teenagers may look at a popular reggae musician as their role model, a case which may not apply to adults who may treasure gospel music. If you can go ahead to ask these teenagers why they treasure reggae, they would give a reason that reggae music has lots of teachings like culture, lifestyle a reason which may not satisfy adults who are hitched to gospel music.

Everyone in society is a role model in his or her own capacity and others are seen as role models by others in society yet they are not aware, though others could be aware. There are two types of role models, positive ethical role models, and negative role models. Parents, religious leaders, political leaders can either be positive or negative role models depending on their behaviors too which can be disastrous or developmental to other actors in the society be it, children or fellow adults.

So the critical issue here is to really see if these role models influence behavior and how. As I have shown in the previous role models influence, both children and adults, and also these role models can be both old people and young people.

These models also vary in that we have religious leaders, social leaders, professional models, personal friends, relatives. A good driving example is Kofi Annan who is considered a professional model because of his tireless efforts to resolve the Kenyan post-election crisis which he succeeded in restoring peace back to Kenya. He is a role model not only to the Kenyan people but also to many other countries which had the same problems but he managed to resolve the crisis. Being retired from a UN secretary post, he only offers services in conflict resolution.

Role models are influential in shaping people’s characters-making these people internalize the ethical behavior they learn, imitate, or copy from their role models. Research on role models and socialization of human beings indicate that role models have great impacts on shaping ethical behavior we see in the world today, for example, parents can shape their children’s behaviors in that the child will exactly copy what he or she sees the father or mother doing making such behavior become part and parcel of the child. Thus role models highly influence and convey ethical standards to young people whether they are wrong or right.

For elementary students, parents also act as their role models. They tend to take their characteristics which later are changed by their attitudes and shaped differently depending on their new friends and associates. For instance, their interaction with their workmates, classmates, and managers will tend to change their ethical behavior and characteristics. It will take an elementary child great tactics to learn from their parents or their surroundings.

For example, if a parent’s way is to make false promises, ever fighting mum when some things go wrong, idle and disorder, then the child is also likely to take ever such characteristics. This makes me believe that however much role model is seen as a positive word, other role models are completely on the negative side and to some extremes are seen as misfits in the society because they just cause trouble instead of living in harmony and bringing positive interactions amongst societal members.

There are many factors affecting the development of behavior of individuals all these sources state that behavior is influenced through imitations which can become permanent. Again Bandura, 1997, children who watch violent movies or for that matter action movies tend to behave in more aggressive ways towards their peers because what they watch remains in their minds, and always want to practice that. This basically shows that if a role model has antisocial or pro-social behaviors, children will definitely copy those behaviors and their social ethics changed in that direction. It is not only children who imitate the behaviors of people they consider to be their role models but even the adults do. These behaviors sometimes become part of individuals hence making them not able to live without them.

Addictive behavior is a behavior that has become part and parcel of one’s character and he or she feels incomplete before actually fulfilling. Just like role models, addictive behavior can either be positive or negative. Addictive behavior comes from within and in most cases, it may not be the same that is being copied or imitated presently but something that was copied in the past could be from a role model and is now just being implemented by another character. Positive addiction can include habitual prayers before sleeping or eating, simple words like thank you, am sorry, please. Negative addiction can include chain-smoking, sexual maniacs, fighting, gossiping.

In the current study, teens with favorable outcome expectations that were balanced by high parental expectations for their behavior were mostly less likely to start drinking than teens with favorable drinking expectations for their behavior. Hence parents should build up their expectations during early adolescents just when teens begin forming their expectations about drinking and still holding their parents as very important to them. While most parents expect their early adolescents not to drink, young people’s perceptions of their parent’s expectations about their behavior may vary depending on the amounts and quality of time parents spend with their children and also the quality and nature of their communication about their expectations on their children’s behavior.

Psychology basically deals with the minds and emotions of individuals. Thus psychology of addictive behaviors is concerned with mental addiction. They are behaviors basically brought about in the mind that is one does what he or she is thinking or has been thinking. Just like the previous two discussed issues, psychology addictive behavior has its extremes of positive and negative aspects which definitely mislead or build up someone. Positive behavior can for instance be courtesy, submission, love while negative behavior can be like pride that is thinking one is better than the rest, hatred bringing negative attitudes towards others. All these behaviors come from within, that is in the mind.

Now the role models here are supposed to act as a big example to the rest. Just like as indicated before, these role models can be parents, politicians, church leaders. Let’s take an example of a church leader as a role model, he or she can help in controlling the psychological addictive behavior of the young and the old in the society more so the positive psychological addictive behaviors. When for instance he or she is going to officiate a church weeding between two lovebirds, he or she will try to bring them to an understanding of why it is important for a wife to submit to a husband and the husband to love his wife by reading to them such scriptures in the bible. He will tell them the advantage and disadvantages of doing so and not doing so.

When counseling these two people they will tend to psychologically adhere to this, thus making them get used to such lifestyle (behavior addiction) so as mainly not to face marriage wrath. On the other hand psychology of addictive behavior can be negative. A good example is that most children look at their parents as role models and if these same parents go astray somewhere they tend to mislead their children. A child who has been brought up in a family where violence is a daily routine cannot be calm in the mind. This child grows up knowing that the only way to solve a problem is not by sitting down and reasoning or having a dialogue but by violence.

This behavior then becomes part and parcel of this child which in further extremes can cause him or her mental depression especially in cases where the child cannot succeed even though he or she uses his or her might. When it reaches such a case, of lets say mental depression, role models like counselors are seen as very important to society, and without them, something is seen to be lacking in society.

The question of whether one is worth living is uprightly important. Human beings act as role models to each other as already mentioned before and have some attachments to the community he lives in. it is said in a proverb that “you may never realize that you have something good until when you lose it” it’s the same here that the importance of human being who are perceived as misfits in the society may not be realized until when they for instance leave or die.

So life is important and everyone is worth living because each person is complex in his or her own ways and thus each person contributes to the society differently and in a special way which can transform lives to the better side which is already seen as role models to others. As long as they are not feeling sick, hungry, and uncomfortable, they will always fill the adult’s atmosphere with humor. This shows that however much someone is seen to be young, but has life; he or she can influence others in a way. Who does not like happiness or humor? Hence these people are seen as role models who inflict happiness in other people’s lives.

So life is an important factor influencing both the role models and those imitating the role models. Remember the dead have got no influence on others, they are just there. Yet life is a conditional function of creatures that influence others.

This then brings up the issue of socialization. Only and only living things socialize. This still shows the importance of life as shown previously. So we must first get to understand the meaning of socialization. In a layman’s understanding, socialization is to take part in social; activities or to take control or adapt to the needs of the society. No society is fully satisfied with what it has because human beings’ needs are insatiable and they keep recurring be it social needs, psychological nits. Economical needs and many others. In this case we are basically going to talk, about the aims of socialization in influencing people’s behavior, more so through role models.

As we have seen before, everyone is a role model in the society thus socialization is entitled to everyone. Without this socialization, most fields in role modeling would not succeed. For example, parents, church leaders, counselors, teachers, peers, husbands and wives need to socialize so, as to succeed as role models. Lets see it in a living example, Let’sough socialization African counselor has succeeded in pulling out street children from the streets taking them to children’s homes or even introducing them to blue-collar jobs like carpentry, masonry which has helped in changing their status in life.

Through socialization, parents have talked telling them the importance of education which has made most children succeed in their lives and eventually have lived comfortably. Church leaders as role models have socialized with their church members through preaching and telling them the importance of sex in marriage and not premarital sexual behaviors. This has helped society in reducing the rate of AIDS scourge, especially among young people thus restoring and maintaining many lives which life is also an important aspect in society for role modeling.

Husbands and wives have also acted as role models to each other in that wives have really shown submission to their husbands thus making them love their wives and hence maintaining marriages and priming a society that is not promiscuous. Even peers have also acted as role models through socialization in schools by encouraging weak colleagues who feel they were not blessed with brains like others. This they do in group discussion thus helping the weak to excel.

Socialization can either be positive or negative, the examples shown above or negative socialization. Negative socialization mostly misleads for example a father who takes his child to a tavern or a pub when taking alcohol with his friends and where obscene words are used is exposing the child to negative socialization because this child can end up becoming an alcoholic, immoral in talks and maybe in deeds. Therefore we realize that socialization is also important when it comes to role modeling.

Socialization bridges the gap between the two extremes of different categories of societal members or even the same categories. This sometimes is seen as difficult without a role model as either initiator or a mediator. Take for example in our day-to-day lives people with high status (the rich) naturally will tend to segregate themselves with those of low status (the poor). The rich will live in posh places while the poor will mainly be found living in slums or average areas.

These two groups will rarely interact or socialize because they believe their needs and wants can never be the same. Little do they know that each and every human being is dependent on the other despite their sex, education, economic, social, and even psychological status. For instance, a poor man’s son can marry a rich man’s daughter and they live comfortably, a rich man’s son can assist a poor man’s daughter in a subject she doesn’t understand and help her excel in education thus making her also rich in future. Here is where the role models in the society try to bridge gaps between people enhancing socialization and thus promoting the spirit of acceptance.

To understand society clearly socialization may be there. This will help to attend to any calamity that may befall society. Human beings being interwoven or dependent units in the society need each other as the saying goes ‘no man is an island.

This can only be accomplished through socialization. No one knows everything that is why even various governments have to involve the grassroots members so as to solve grass root problems. This is done through a bottom-up approach. On the contrary also for the government to succeed in its service provision after getting to know grass root problems it has to use a top-down approach. Top-down and bottom-up approaches are two ways of socialization which mostly is enhanced by role models who are sincere.

Through socialization native language and culture are passed with their originality. In this case, the best role models here are our grandparents or for that matter our forefathers. Our grandparents also are seen only being able to pass on the native language and culture through socialization. The old people socialize by telling stories, teaching cultural practices for instance in Africa many grandmothers teach their granddaughters how to take care of a husband, how to cook the native local foods, how to take care of their babies after giving birth while grandfathers are seen to be teaching their grandsons how to take care of their homes, how to hunt and bring food home, how to command respect from their families and how to treat their wives.

The fourth aim of socialization is to maintain societal norms, virtues, and beliefs. The church leaders, government arms are good examples here because they act as role models to maintain these. Church leader’s role models will inflict virtues like telling people the importance of living in harmony with for instance stealing, killing, smoking, irresponsible sexual behaviors. These they only do through socialization which eventually can transform some societal members. So as we have seen in all these points above, socialization aims at transforming society for the better and the main transformers are the role models who only interact or socialize with others when trying to influence them positively.

Socialization being a social activity must have a process for its smooth running. This process of activities must start from somewhere going step by step in order for instance when two people meet and fall in love and even plan to marry, they will first get to know each other name, age, origin, and even denomination for some people. From that point, they will become normal friends with no ultimate socialization than they later become intimate and start courting and eventually tie the knots thus becoming husband and wife. In their process of socialization, they will have their role models like God the almighty, other friends who may be married, and living comfortably to help them go about this issue correctly.

When babies are born, they totally know no one and nothing. As they grow, they learn things around them, who their parents are, when they are hungry, when they have been left alone. Some qualities may vary from child to child depending on their role models who are mainly their parents. Within the environment in which the child is born, he or she learns the language, customs, attitudes, and various things in which things are done in his or her society. Therefore different environments will play different vital parts in molding the potentiality and effect they learn to behave.

I have seen everyone as a role model in society in his or her own ways. Role models have also been portrayed as either positive influencers or negative influencers. Role models have also been seen as having vast professions or status in society. Like teachers, church leaders, government leaders, grandparents, peers, parents, husbands and wives, sisters, brothers, the rich, and the poor.

Role models have been seen as actors who shape people’s characters where others become addictive behaviors that are either positive or negative. Role models are also seen to be great actors in psychological addictive behaviors. For instance, counselors, as role models have tried to advise let’s, say street children to move away from the streets a place where these children have gotten used to and are psychologically comfortable there.

Socialization is also an issue that has sheen discussed her stating that its why only through socialization that the role models can be able to convey their influences to others who an example given is grandparents teaching their grandchildren native cultures and language through socialization. Human life as an important aspect in society has also promoted role models influences to other people. Here I have seen that there is no man who is an island.

The socialization process is important because it helps the role models take for instance counselors advise two people wishing to marry on what to expect in marriage and how to handle these issues.

In conclusion, it can be seen that these role models are very influential and in most instances important to society and that’s why society needs them.

Bandura A. (1971) Social learning theory General Learning . Press: New York.

Bandura A. (1974) ‘Behaviour theory and the models of man’, American Psychologist.

Bandura A. (1997), The Exercise of Control .

Bandura A. (1977) Self-Efficacy: Toward a Unifying Theory of Behavioral Change , 84(2) Psychological Review 191-215.

Chicago (A-D)
Chicago (N-B)

IvyPanda. (2021, September 2). How Role Models Influence Behaviors. https://ivypanda.com/essays/how-role-models-influence-behaviors/

"How Role Models Influence Behaviors." IvyPanda , 2 Sept. 2021, ivypanda.com/essays/how-role-models-influence-behaviors/.

IvyPanda . (2021) 'How Role Models Influence Behaviors'. 2 September.

IvyPanda . 2021. "How Role Models Influence Behaviors." September 2, 2021. https://ivypanda.com/essays/how-role-models-influence-behaviors/.

1. IvyPanda . "How Role Models Influence Behaviors." September 2, 2021. https://ivypanda.com/essays/how-role-models-influence-behaviors/.

Bibliography

IvyPanda . "How Role Models Influence Behaviors." September 2, 2021. https://ivypanda.com/essays/how-role-models-influence-behaviors/.

The Psychology of Addictive Behavior
Addictive Personality in Drug Rehabilitation
Mood and Addictive Disorders in Psychology
Addictive Behaviours Types Analysis
Methamphetamine Is Different and Addictive
The Psychology of Addiction and Addictive Behaviors
Compulsive and Addictive Behaviors
Heroin Addictive Behavior and Its Etiology
Drug Abuse: Age, Gender and Addictive Susceptibility
Excessive Use of Social Media and Addictive Behavior
Human Experiential Behavior and Love Relationship
Aggression and the Role of Brain
The Act of Committing Suicide Issue
Married Couples' Behavior Change When Observed
Behavioral Self-Control and Its Factors

Environment
Information Science
Social Issues
Argumentative
Cause and Effect
Classification
Compare and Contrast
Descriptive
Exemplification
Informative
Controversial
Exploratory
What Is an Essay
Length of an Essay
Generate Ideas
Types of Essays
Structuring an Essay
Outline For Essay
Essay Introduction
Thesis Statement
Body of an Essay
Writing a Conclusion
Essay Writing Tips
Drafting an Essay
Revision Process
Fix a Broken Essay
Format of an Essay
Essay Examples
Essay Checklist
Essay Writing Service
Pay for Research Paper
Write My Research Paper
Write My Essay
Custom Essay Writing Service
Admission Essay Writing Service
Pay for Essay
Academic Ghostwriting
Write My Book Report
Case Study Writing Service
Dissertation Writing Service
Coursework Writing Service
Lab Report Writing Service
Do My Assignment
Buy College Papers
Capstone Project Writing Service
Buy Research Paper
Custom Essays for Sale

Can’t find a perfect paper?

Free Essay Samples
Entertainment
Role Models

Essays on Role Models

Most people in their life at some point had to write a role models essay. The concept of role models is ancient. When we read a book or watch a movie, we often gravitate towards a particular character. We can't help but root for them, we emphasize with them, relate to their problems, share their values, we want to be their friend – ultimately, they become our role model. Sounds relatable? Such characters are designed to attract people – they usually have a strong moral compass, are honest, honorable, selfless, or have other distinctive good qualities. Real-life people can also become role models, usually due to their contribution to the world. Role models essays are generally about people who we respect and want to resemble. Our role models essay samples feature many essays about remarkable people. Feel free to check out these samples of essays on role models, as they might help enrich your own essay.

Since childhood, I have acquired a number of heroes in my life depending on their role in my life. My first heroes were my parents who brought me up and ensured I had everything I needed. Moreover, being career people, they served as a source of motivation for the person I...

Condoleezza Rice is a US diplomat who served as the 66th US Secretary of State and is widely regarded as a role model for other African American women aspiring to professional success. Rice was the only child born to Angelena and John Wesley Rice, Jr. Angelena worked as a teacher...

Words: 3667

Found a perfect essay sample but want a unique one?

Request writing help from expert writer in you feed!

It is important for individuals who advance to positions of leadership, regardless of title, to have a clear understanding of what leadership entails. Beyond simply being a figure who wields authority, it is critical for individuals to exhibit reflective behavior that is appropriate for the role they occupy. Leaders bear...

The Ultimate Role Model: Jesus The ultimate model is Jesus for Christians; he is the individual whose personality, actions, example and norm all can imite (Blanchard and Phil 18). According to the Bible, "Because God had previously known his people, he chose them as his Son, so that he would make...

Creative samples from the experts

↑ Return to Essay Samples

Descriptive Essay: My Role Models

Many people look to someone to serve as a model or motivator for their life. They want someone to look up to, who has good values and a sense of honor and integrity. Most people choose one person to act as their ‘guiding light’ for their life, but I am slightly different. I have chosen several. No one is perfect, after all, so I have a few people whose traits or deeds I admire. I think of them often and hope to live my life in a similar way.

My first role model has been with me since I saw his bronzed shoes in a museum when I was five. The shoes belong to Terry Fox. He’s a hero to an entire nation because he raced across Canada with one artificial leg in order to raise money for cancer research. He endured horrible weather, illness, physical pain, and other extreme hardships in order to meet his goal. I admire him for what he attempted, even though he did not succeed. Terry Fox had wanted to race across Canada, but did not make it more than a third of the way. He’s famous now, but was unknown at the time, so what he did was mostly in anonymity. But he never gave up. He had a goal and he worked toward it, no matter the end result.

My second role model is my best friend Rachel. I chose her because she is close to me. She is someone I can connect with on a daily basis, and watch as she muddles through life’s daily tribulations with as much honor and respect as she can muster. Has she undergone incredible hardships in her life? No. Not really. But she lives through the politics of every day adult life trying to be honest with herself and with others, avoiding gossip and being generous with people. Every time she buys me dinner, or spends time with my young niece, I’m reminded that people can be generous and caring.

The last person I look to as a role model is myself. I find that I do not always have to look outside myself for motivation and positivity. I have endured hardships in my life. I’ve lost family members to illness and stood up to defend myself from abuse. I live with an anxiety disorder and have to continually remind myself about various calming techniques. But I still stay positive and do my best every day. I can look back at things I’ve accomplished in my life and be proud. I can look at what I’ve overcome and realize that I am strong.

Role models are important. Life can be hard, with money woes and illness dragging us down. People need tangible reminders of the fact that strength and honor can triumph in rough circumstance. They need to be aware that every day people can be generous and caring. And people need to know that they, themselves, are brave and strong.

Follow Us on Social Media

Get more free essays

Send via email

Most useful resources for students:.

Free Essays Download
Writing Tools List
Proofreading Services
Universities Rating

Contributors Bio

Find more useful services for students

Free plagiarism check, professional editing, online tutoring, free grammar check.

CBSE Class 10th
CBSE Class 12th
UP Board 10th
UP Board 12th
Bihar Board 10th
Bihar Board 12th
Top Schools in India
Top Schools in Delhi
Top Schools in Mumbai
Top Schools in Chennai
Top Schools in Hyderabad
Top Schools in Kolkata
Top Schools in Pune
Top Schools in Bangalore

Products & Resources

JEE Main Knockout April
Free Sample Papers
Free Ebooks
NCERT Notes
NCERT Syllabus
NCERT Books
RD Sharma Solutions
Navodaya Vidyalaya Admission 2024-25
NCERT Solutions
NCERT Solutions for Class 12
NCERT Solutions for Class 11
NCERT solutions for Class 10
NCERT solutions for Class 9
NCERT solutions for Class 8
NCERT Solutions for Class 7
JEE Main 2024
JEE Advanced 2024
BITSAT 2024
View All Engineering Exams
Colleges Accepting B.Tech Applications
Top Engineering Colleges in India
Engineering Colleges in India
Engineering Colleges in Tamil Nadu
Engineering Colleges Accepting JEE Main
Top IITs in India
Top NITs in India
Top IIITs in India
JEE Main College Predictor
JEE Main Rank Predictor
MHT CET College Predictor
AP EAMCET College Predictor
GATE College Predictor
KCET College Predictor
JEE Advanced College Predictor
View All College Predictors
JEE Main Question Paper
JEE Main Mock Test
JEE Main Registration
JEE Main Syllabus
Download E-Books and Sample Papers
Compare Colleges
B.Tech College Applications
GATE 2024 Result
MAH MBA CET Exam
View All Management Exams

Colleges & Courses

MBA College Admissions
MBA Colleges in India
Top IIMs Colleges in India
Top Online MBA Colleges in India
MBA Colleges Accepting XAT Score
BBA Colleges in India
XAT College Predictor 2024
SNAP College Predictor
NMAT College Predictor
MAT College Predictor 2024
CMAT College Predictor 2024
CAT Percentile Predictor 2023
CAT 2023 College Predictor
CMAT 2024 Registration
TS ICET 2024 Registration
CMAT Exam Date 2024
MAH MBA CET Cutoff 2024
Download Helpful Ebooks
List of Popular Branches
QnA - Get answers to your doubts
IIM Fees Structure
AIIMS Nursing
Top Medical Colleges in India
Top Medical Colleges in India accepting NEET Score
Medical Colleges accepting NEET
List of Medical Colleges in India
List of AIIMS Colleges In India
Medical Colleges in Maharashtra
Medical Colleges in India Accepting NEET PG
NEET College Predictor
NEET PG College Predictor
NEET MDS College Predictor
DNB CET College Predictor
DNB PDCET College Predictor
NEET Application Form 2024
NEET PG Application Form 2024
NEET Cut off
NEET Online Preparation
Download Helpful E-books
LSAT India 2024
Colleges Accepting Admissions
Top Law Colleges in India
Law College Accepting CLAT Score
List of Law Colleges in India
Top Law Colleges in Delhi
Top Law Collages in Indore
Top Law Colleges in Chandigarh
Top Law Collages in Lucknow

Predictors & E-Books

CLAT College Predictor
MHCET Law ( 5 Year L.L.B) College Predictor
AILET College Predictor
Sample Papers
Compare Law Collages
Careers360 Youtube Channel
CLAT Syllabus 2025
CLAT Previous Year Question Paper
AIBE 18 Result 2023
NID DAT Exam
Pearl Academy Exam

Animation Courses

Animation Courses in India
Animation Courses in Bangalore
Animation Courses in Mumbai
Animation Courses in Pune
Animation Courses in Chennai
Animation Courses in Hyderabad
Design Colleges in India
Fashion Design Colleges in Bangalore
Fashion Design Colleges in Mumbai
Fashion Design Colleges in Pune
Fashion Design Colleges in Delhi
Fashion Design Colleges in Hyderabad
Fashion Design Colleges in India
Top Design Colleges in India
Free Design E-books
List of Branches
Careers360 Youtube channel
NIFT College Predictor
UCEED College Predictor
NID DAT College Predictor
IPU CET BJMC
JMI Mass Communication Entrance Exam
IIMC Entrance Exam
Media & Journalism colleges in Delhi
Media & Journalism colleges in Bangalore
Media & Journalism colleges in Mumbai
List of Media & Journalism Colleges in India
CA Intermediate
CA Foundation
CS Executive
CS Professional
Difference between CA and CS
Difference between CA and CMA
CA Full form
CMA Full form
CS Full form
CA Salary In India

Top Courses & Careers

Bachelor of Commerce (B.Com)
Master of Commerce (M.Com)
Company Secretary
Cost Accountant
Charted Accountant
Credit Manager
Financial Advisor
Top Commerce Colleges in India
Top Government Commerce Colleges in India
Top Private Commerce Colleges in India
Top M.Com Colleges in Mumbai
Top B.Com Colleges in India
IT Colleges in Tamil Nadu
IT Colleges in Uttar Pradesh
MCA Colleges in India
BCA Colleges in India

Diploma Colleges

Top Diploma Colleges in Maharashtra
UPSC IAS 2024
SSC CGL 2024
IBPS RRB 2024
Previous Year Sample Papers
Free Competition E-books
Sarkari Result
QnA- Get your doubts answered
UPSC Previous Year Sample Papers
CTET Previous Year Sample Papers
SBI Clerk Previous Year Sample Papers
NDA Previous Year Sample Papers

Upcoming Events

NDA Application Form 2024
UPSC IAS Application Form 2024
CDS Application Form 2024
CTET Admit card 2024
HP TET Result 2023
SSC GD Constable Admit Card 2024
UPTET Notification 2024
SBI Clerk Result 2024

Other Exams

SSC CHSL 2024
UP PCS 2024
UGC NET 2024
RRB NTPC 2024
IBPS PO 2024
IBPS Clerk 2024
IBPS SO 2024
Top University in USA
Top University in Canada
Top University in Ireland
Top Universities in UK
Top Universities in Australia
Best MBA Colleges in Abroad
Business Management Studies Colleges

Top Countries

Study in USA
Study in UK
Study in Canada
Study in Australia
Study in Ireland
Study in Germany
Study in China
Study in Europe

Student Visas

Student Visa Canada
Student Visa UK
Student Visa USA
Student Visa Australia
Student Visa Germany
Student Visa New Zealand
Student Visa Ireland
CUET PG 2024
IGNOU B.Ed Admission 2024
DU Admission
UP B.Ed JEE 2024
DDU Entrance Exam
IIT JAM 2024
IGNOU Online Admission 2024
Universities in India
Top Universities in India 2024
Top Colleges in India
Top Universities in Uttar Pradesh 2024
Top Universities in Bihar
Top Universities in Madhya Pradesh 2024
Top Universities in Tamil Nadu 2024
Central Universities in India
CUET PG Admit Card 2024
IGNOU Date Sheet
CUET Mock Test 2024
CUET Application Form 2024
CUET PG Syllabus 2024
CUET Participating Universities 2024
CUET Previous Year Question Paper
CUET Syllabus 2024 for Science Students
E-Books and Sample Papers
CUET Exam Pattern 2024
CUET Exam Date 2024
CUET Syllabus 2024
IGNOU Exam Form 2024
IGNOU Result
CUET PG Courses 2024

Engineering Preparation

Knockout JEE Main 2024
Test Series JEE Main 2024
JEE Main 2024 Rank Booster

Medical Preparation

Knockout NEET 2024
Test Series NEET 2024
Rank Booster NEET 2024

Online Courses

JEE Main One Month Course
NEET One Month Course
IBSAT Free Mock Tests
IIT JEE Foundation Course
Knockout BITSAT 2024
Career Guidance Tool

Top Streams

IT & Software Certification Courses
Engineering and Architecture Certification Courses
Programming And Development Certification Courses
Business and Management Certification Courses
Marketing Certification Courses
Health and Fitness Certification Courses
Design Certification Courses

Specializations

Digital Marketing Certification Courses
Cyber Security Certification Courses
Artificial Intelligence Certification Courses
Business Analytics Certification Courses
Data Science Certification Courses
Cloud Computing Certification Courses
Machine Learning Certification Courses
View All Certification Courses
UG Degree Courses
PG Degree Courses
Short Term Courses
Free Courses
Online Degrees and Diplomas
Compare Courses

Top Providers

Coursera Courses
Udemy Courses
Edx Courses
Swayam Courses
upGrad Courses
Simplilearn Courses
Great Learning Courses

Access premium articles, webinars, resources to make the best decisions for career, course, exams, scholarships, study abroad and much more with

Plan, Prepare & Make the Best Career Choices

My Role Model Essay

Role models are the people to which we look upon as a source of inspiration and motivation. Many people inspire us, but very few are our role models. Here are a few sample essays on the topic ‘my role model’.

100 Words Essay on My Role Model

The one who gave birth to me is my role model. The amount of sacrifice and pain she had borne is unpayable. I look upon her as the strongest woman and someone who is a statue of morality. I always try to follow her guidelines and ideals. My mother is a living goddess, who always works for the well-being of others. She never complains to anyone about her troubles. All she wants is love and respect. I admire her behavior and deeds in my life. I aim to work as hard as her wish to make her feel proud of me.

200 Words Essay on My Role Model

500 words essay on my role model.

Sushmita Sen is my role model. Her confident and fearless behavior makes me ardently follow her. The speeches that she delivers on various platforms or the interviews she gives, reflect upon her thoughts that are magical and powerful, and she has inspired many girls and women across the world. I aspire to become a confident woman like her. I truly admire her as a person. She is one of the most beautiful women in the world. She is a feminist icon who supports and encourages women.

She is an outstanding actor and model. In the year 1994, she became Miss Universe at the age of 18, making all the Indians proud. Her most vigorous decision was to adopt two girl children as a single mother. Many hurdles came to her while adopting them, but she fluently passed all the hurdles for her girls.

Many denied her from adopting a child as a single mother. She fought the whole world for both of her daughters. She stood like a strong tree showing that small winds cannot break her. She taught us to believe in our dreams and work hard to accomplish them. She is a single mother showing the world that a woman is capable and independent of raising her children independently.

A role model is a person who can inspire people. Ratan Tata is the most generous person on earth and is my role model. Despite being the chairman of Tata group of industries, he is known for his golden heart. He is a man with high values. His motivating speech and quotes manifest his thoughts.

He is a philanthropist who donates a large amount of money to charity. His pure heart and dedication are the reason for the tremendous success of Tata industries. A visionary leader who motivates his employees to work towards a new vision and new opportunities. I look upon him as a person full of values, a risk-taker, and a man with a purely positive attitude. He is a role model for the youth.

Qualities of My Role Model

Ratan Tata is a humble man who insists on not getting special treatment. He drives his car and lives a simple life. He is my ideal and believes in kindness, simplicity, and empathy. He is an 84-year-old industrialist who has consistently demonstrated that he is much more than a business person. I want to be charitable and selfless like him. He inspired many people to take a step towards charity.

Inspiring Quotes by Ratan Tata

Ratan Tata is a humble man whose words and quotes resonate with me and inspire me to be a better person aiming for success and peace. These are some of my favourite quotes by him—

‘None can destroy iron, but its own rust can. Likewise, none can destroy a person but his own mindset can’—This powerful quote suggests that nothing can stop a person from being successful, but only his mindset.

‘Take the stones, people, to throw at you and use them to build a monument’—This great thought inspires people to do what they want. No matter what others are saying or what others are accomplishing.

Reasons Why He is a Role Model

Sometimes, he travels on an economy class ticket for flights and surprises his fellow passengers. This shows that he is not a man who lives a luxurious life and teaches us that no matter how rich we are, we must stay grounded.

His primary focus is on the severe issue of malnutrition. He said that the health of children and pregnant women is essential as it will change the mental and physical health of the nation. He believes that children are the nation's future, and we need to nourish children for a bright future for the country.

During the COVID-19 pandemic, he recommended companies not lay off their employees. Also, during the pandemic, he donated hundreds of crores to the nation for the help of frontline workers and those who are fighting the virus.

He believes that everyone has the right to education and lack of money should not be a hurdle.

He visited around 80 families who were impacted by the 26/11 attack. He helped many street dogs and also requested his followers to do so. He always inspires and motivates people on his social media handles.

Explore Career Options (By Industry)

Construction
Entertainment
Manufacturing
Information Technology

Data Administrator

Database professionals use software to store and organise data such as financial information, and customer shipping records. Individuals who opt for a career as data administrators ensure that data is available for users and secured from unauthorised sales. DB administrators may work in various types of industries. It may involve computer systems design, service firms, insurance companies, banks and hospitals.

Bio Medical Engineer

The field of biomedical engineering opens up a universe of expert chances. An Individual in the biomedical engineering career path work in the field of engineering as well as medicine, in order to find out solutions to common problems of the two fields. The biomedical engineering job opportunities are to collaborate with doctors and researchers to develop medical systems, equipment, or devices that can solve clinical problems. Here we will be discussing jobs after biomedical engineering, how to get a job in biomedical engineering, biomedical engineering scope, and salary.

Ethical Hacker

A career as ethical hacker involves various challenges and provides lucrative opportunities in the digital era where every giant business and startup owns its cyberspace on the world wide web. Individuals in the ethical hacker career path try to find the vulnerabilities in the cyber system to get its authority. If he or she succeeds in it then he or she gets its illegal authority. Individuals in the ethical hacker career path then steal information or delete the file that could affect the business, functioning, or services of the organization.

GIS officer work on various GIS software to conduct a study and gather spatial and non-spatial information. GIS experts update the GIS data and maintain it. The databases include aerial or satellite imagery, latitudinal and longitudinal coordinates, and manually digitized images of maps. In a career as GIS expert, one is responsible for creating online and mobile maps.

Data Analyst

The invention of the database has given fresh breath to the people involved in the data analytics career path. Analysis refers to splitting up a whole into its individual components for individual analysis. Data analysis is a method through which raw data are processed and transformed into information that would be beneficial for user strategic thinking.

Data are collected and examined to respond to questions, evaluate hypotheses or contradict theories. It is a tool for analyzing, transforming, modeling, and arranging data with useful knowledge, to assist in decision-making and methods, encompassing various strategies, and is used in different fields of business, research, and social science.

Geothermal Engineer

Individuals who opt for a career as geothermal engineers are the professionals involved in the processing of geothermal energy. The responsibilities of geothermal engineers may vary depending on the workplace location. Those who work in fields design facilities to process and distribute geothermal energy. They oversee the functioning of machinery used in the field.

Database Architect

If you are intrigued by the programming world and are interested in developing communications networks then a career as database architect may be a good option for you. Data architect roles and responsibilities include building design models for data communication networks. Wide Area Networks (WANs), local area networks (LANs), and intranets are included in the database networks. It is expected that database architects will have in-depth knowledge of a company's business to develop a network to fulfil the requirements of the organisation. Stay tuned as we look at the larger picture and give you more information on what is db architecture, why you should pursue database architecture, what to expect from such a degree and what your job opportunities will be after graduation. Here, we will be discussing how to become a data architect. Students can visit NIT Trichy , IIT Kharagpur , JMI New Delhi .

Remote Sensing Technician

Individuals who opt for a career as a remote sensing technician possess unique personalities. Remote sensing analysts seem to be rational human beings, they are strong, independent, persistent, sincere, realistic and resourceful. Some of them are analytical as well, which means they are intelligent, introspective and inquisitive.

Remote sensing scientists use remote sensing technology to support scientists in fields such as community planning, flight planning or the management of natural resources. Analysing data collected from aircraft, satellites or ground-based platforms using statistical analysis software, image analysis software or Geographic Information Systems (GIS) is a significant part of their work. Do you want to learn how to become remote sensing technician? There's no need to be concerned; we've devised a simple remote sensing technician career path for you. Scroll through the pages and read.

Budget Analyst

Budget analysis, in a nutshell, entails thoroughly analyzing the details of a financial budget. The budget analysis aims to better understand and manage revenue. Budget analysts assist in the achievement of financial targets, the preservation of profitability, and the pursuit of long-term growth for a business. Budget analysts generally have a bachelor's degree in accounting, finance, economics, or a closely related field. Knowledge of Financial Management is of prime importance in this career.

Underwriter

An underwriter is a person who assesses and evaluates the risk of insurance in his or her field like mortgage, loan, health policy, investment, and so on and so forth. The underwriter career path does involve risks as analysing the risks means finding out if there is a way for the insurance underwriter jobs to recover the money from its clients. If the risk turns out to be too much for the company then in the future it is an underwriter who will be held accountable for it. Therefore, one must carry out his or her job with a lot of attention and diligence.

Finance Executive

Product manager.

A Product Manager is a professional responsible for product planning and marketing. He or she manages the product throughout the Product Life Cycle, gathering and prioritising the product. A product manager job description includes defining the product vision and working closely with team members of other departments to deliver winning products.

Operations Manager

Individuals in the operations manager jobs are responsible for ensuring the efficiency of each department to acquire its optimal goal. They plan the use of resources and distribution of materials. The operations manager's job description includes managing budgets, negotiating contracts, and performing administrative tasks.

Stock Analyst

Individuals who opt for a career as a stock analyst examine the company's investments makes decisions and keep track of financial securities. The nature of such investments will differ from one business to the next. Individuals in the stock analyst career use data mining to forecast a company's profits and revenues, advise clients on whether to buy or sell, participate in seminars, and discussing financial matters with executives and evaluate annual reports.

A Researcher is a professional who is responsible for collecting data and information by reviewing the literature and conducting experiments and surveys. He or she uses various methodological processes to provide accurate data and information that is utilised by academicians and other industry professionals. Here, we will discuss what is a researcher, the researcher's salary, types of researchers.

Welding Engineer

Welding Engineer Job Description: A Welding Engineer work involves managing welding projects and supervising welding teams. He or she is responsible for reviewing welding procedures, processes and documentation. A career as Welding Engineer involves conducting failure analyses and causes on welding issues.

Transportation Planner

A career as Transportation Planner requires technical application of science and technology in engineering, particularly the concepts, equipment and technologies involved in the production of products and services. In fields like land use, infrastructure review, ecological standards and street design, he or she considers issues of health, environment and performance. A Transportation Planner assigns resources for implementing and designing programmes. He or she is responsible for assessing needs, preparing plans and forecasts and compliance with regulations.

Environmental Engineer

Individuals who opt for a career as an environmental engineer are construction professionals who utilise the skills and knowledge of biology, soil science, chemistry and the concept of engineering to design and develop projects that serve as solutions to various environmental problems.

Safety Manager

A Safety Manager is a professional responsible for employee’s safety at work. He or she plans, implements and oversees the company’s employee safety. A Safety Manager ensures compliance and adherence to Occupational Health and Safety (OHS) guidelines.

Conservation Architect

A Conservation Architect is a professional responsible for conserving and restoring buildings or monuments having a historic value. He or she applies techniques to document and stabilise the object’s state without any further damage. A Conservation Architect restores the monuments and heritage buildings to bring them back to their original state.

Structural Engineer

A Structural Engineer designs buildings, bridges, and other related structures. He or she analyzes the structures and makes sure the structures are strong enough to be used by the people. A career as a Structural Engineer requires working in the construction process. It comes under the civil engineering discipline. A Structure Engineer creates structural models with the help of computer-aided design software.

Highway Engineer

Highway Engineer Job Description: A Highway Engineer is a civil engineer who specialises in planning and building thousands of miles of roads that support connectivity and allow transportation across the country. He or she ensures that traffic management schemes are effectively planned concerning economic sustainability and successful implementation.

Field Surveyor

Are you searching for a Field Surveyor Job Description? A Field Surveyor is a professional responsible for conducting field surveys for various places or geographical conditions. He or she collects the required data and information as per the instructions given by senior officials.

Orthotist and Prosthetist

Orthotists and Prosthetists are professionals who provide aid to patients with disabilities. They fix them to artificial limbs (prosthetics) and help them to regain stability. There are times when people lose their limbs in an accident. In some other occasions, they are born without a limb or orthopaedic impairment. Orthotists and prosthetists play a crucial role in their lives with fixing them to assistive devices and provide mobility.

Pathologist

A career in pathology in India is filled with several responsibilities as it is a medical branch and affects human lives. The demand for pathologists has been increasing over the past few years as people are getting more aware of different diseases. Not only that, but an increase in population and lifestyle changes have also contributed to the increase in a pathologist’s demand. The pathology careers provide an extremely huge number of opportunities and if you want to be a part of the medical field you can consider being a pathologist. If you want to know more about a career in pathology in India then continue reading this article.

Veterinary Doctor

Speech therapist, gynaecologist.

Gynaecology can be defined as the study of the female body. The job outlook for gynaecology is excellent since there is evergreen demand for one because of their responsibility of dealing with not only women’s health but also fertility and pregnancy issues. Although most women prefer to have a women obstetrician gynaecologist as their doctor, men also explore a career as a gynaecologist and there are ample amounts of male doctors in the field who are gynaecologists and aid women during delivery and childbirth.

Audiologist

The audiologist career involves audiology professionals who are responsible to treat hearing loss and proactively preventing the relevant damage. Individuals who opt for a career as an audiologist use various testing strategies with the aim to determine if someone has a normal sensitivity to sounds or not. After the identification of hearing loss, a hearing doctor is required to determine which sections of the hearing are affected, to what extent they are affected, and where the wound causing the hearing loss is found. As soon as the hearing loss is identified, the patients are provided with recommendations for interventions and rehabilitation such as hearing aids, cochlear implants, and appropriate medical referrals. While audiology is a branch of science that studies and researches hearing, balance, and related disorders.

An oncologist is a specialised doctor responsible for providing medical care to patients diagnosed with cancer. He or she uses several therapies to control the cancer and its effect on the human body such as chemotherapy, immunotherapy, radiation therapy and biopsy. An oncologist designs a treatment plan based on a pathology report after diagnosing the type of cancer and where it is spreading inside the body.

Are you searching for an ‘Anatomist job description’? An Anatomist is a research professional who applies the laws of biological science to determine the ability of bodies of various living organisms including animals and humans to regenerate the damaged or destroyed organs. If you want to know what does an anatomist do, then read the entire article, where we will answer all your questions.

For an individual who opts for a career as an actor, the primary responsibility is to completely speak to the character he or she is playing and to persuade the crowd that the character is genuine by connecting with them and bringing them into the story. This applies to significant roles and littler parts, as all roles join to make an effective creation. Here in this article, we will discuss how to become an actor in India, actor exams, actor salary in India, and actor jobs.

Individuals who opt for a career as acrobats create and direct original routines for themselves, in addition to developing interpretations of existing routines. The work of circus acrobats can be seen in a variety of performance settings, including circus, reality shows, sports events like the Olympics, movies and commercials. Individuals who opt for a career as acrobats must be prepared to face rejections and intermittent periods of work. The creativity of acrobats may extend to other aspects of the performance. For example, acrobats in the circus may work with gym trainers, celebrities or collaborate with other professionals to enhance such performance elements as costume and or maybe at the teaching end of the career.

Video Game Designer

Career as a video game designer is filled with excitement as well as responsibilities. A video game designer is someone who is involved in the process of creating a game from day one. He or she is responsible for fulfilling duties like designing the character of the game, the several levels involved, plot, art and similar other elements. Individuals who opt for a career as a video game designer may also write the codes for the game using different programming languages.

Depending on the video game designer job description and experience they may also have to lead a team and do the early testing of the game in order to suggest changes and find loopholes.

Radio Jockey

Radio Jockey is an exciting, promising career and a great challenge for music lovers. If you are really interested in a career as radio jockey, then it is very important for an RJ to have an automatic, fun, and friendly personality. If you want to get a job done in this field, a strong command of the language and a good voice are always good things. Apart from this, in order to be a good radio jockey, you will also listen to good radio jockeys so that you can understand their style and later make your own by practicing.

A career as radio jockey has a lot to offer to deserving candidates. If you want to know more about a career as radio jockey, and how to become a radio jockey then continue reading the article.

Choreographer

The word “choreography" actually comes from Greek words that mean “dance writing." Individuals who opt for a career as a choreographer create and direct original dances, in addition to developing interpretations of existing dances. A Choreographer dances and utilises his or her creativity in other aspects of dance performance. For example, he or she may work with the music director to select music or collaborate with other famous choreographers to enhance such performance elements as lighting, costume and set design.

Social Media Manager

A career as social media manager involves implementing the company’s or brand’s marketing plan across all social media channels. Social media managers help in building or improving a brand’s or a company’s website traffic, build brand awareness, create and implement marketing and brand strategy. Social media managers are key to important social communication as well.

Photographer

Photography is considered both a science and an art, an artistic means of expression in which the camera replaces the pen. In a career as a photographer, an individual is hired to capture the moments of public and private events, such as press conferences or weddings, or may also work inside a studio, where people go to get their picture clicked. Photography is divided into many streams each generating numerous career opportunities in photography. With the boom in advertising, media, and the fashion industry, photography has emerged as a lucrative and thrilling career option for many Indian youths.

An individual who is pursuing a career as a producer is responsible for managing the business aspects of production. They are involved in each aspect of production from its inception to deception. Famous movie producers review the script, recommend changes and visualise the story.

They are responsible for overseeing the finance involved in the project and distributing the film for broadcasting on various platforms. A career as a producer is quite fulfilling as well as exhaustive in terms of playing different roles in order for a production to be successful. Famous movie producers are responsible for hiring creative and technical personnel on contract basis.

Copy Writer

In a career as a copywriter, one has to consult with the client and understand the brief well. A career as a copywriter has a lot to offer to deserving candidates. Several new mediums of advertising are opening therefore making it a lucrative career choice. Students can pursue various copywriter courses such as Journalism , Advertising , Marketing Management . Here, we have discussed how to become a freelance copywriter, copywriter career path, how to become a copywriter in India, and copywriting career outlook.

In a career as a vlogger, one generally works for himself or herself. However, once an individual has gained viewership there are several brands and companies that approach them for paid collaboration. It is one of those fields where an individual can earn well while following his or her passion.

Ever since internet costs got reduced the viewership for these types of content has increased on a large scale. Therefore, a career as a vlogger has a lot to offer. If you want to know more about the Vlogger eligibility, roles and responsibilities then continue reading the article.

For publishing books, newspapers, magazines and digital material, editorial and commercial strategies are set by publishers. Individuals in publishing career paths make choices about the markets their businesses will reach and the type of content that their audience will be served. Individuals in book publisher careers collaborate with editorial staff, designers, authors, and freelance contributors who develop and manage the creation of content.

Careers in journalism are filled with excitement as well as responsibilities. One cannot afford to miss out on the details. As it is the small details that provide insights into a story. Depending on those insights a journalist goes about writing a news article. A journalism career can be stressful at times but if you are someone who is passionate about it then it is the right choice for you. If you want to know more about the media field and journalist career then continue reading this article.

Individuals in the editor career path is an unsung hero of the news industry who polishes the language of the news stories provided by stringers, reporters, copywriters and content writers and also news agencies. Individuals who opt for a career as an editor make it more persuasive, concise and clear for readers. In this article, we will discuss the details of the editor's career path such as how to become an editor in India, editor salary in India and editor skills and qualities.

Individuals who opt for a career as a reporter may often be at work on national holidays and festivities. He or she pitches various story ideas and covers news stories in risky situations. Students can pursue a BMC (Bachelor of Mass Communication) , B.M.M. (Bachelor of Mass Media) , or MAJMC (MA in Journalism and Mass Communication) to become a reporter. While we sit at home reporters travel to locations to collect information that carries a news value.

Corporate Executive

Are you searching for a Corporate Executive job description? A Corporate Executive role comes with administrative duties. He or she provides support to the leadership of the organisation. A Corporate Executive fulfils the business purpose and ensures its financial stability. In this article, we are going to discuss how to become corporate executive.

Multimedia Specialist

A multimedia specialist is a media professional who creates, audio, videos, graphic image files, computer animations for multimedia applications. He or she is responsible for planning, producing, and maintaining websites and applications.

Quality Controller

A quality controller plays a crucial role in an organisation. He or she is responsible for performing quality checks on manufactured products. He or she identifies the defects in a product and rejects the product.

A quality controller records detailed information about products with defects and sends it to the supervisor or plant manager to take necessary actions to improve the production process.

Production Manager

A QA Lead is in charge of the QA Team. The role of QA Lead comes with the responsibility of assessing services and products in order to determine that he or she meets the quality standards. He or she develops, implements and manages test plans.

Process Development Engineer

The Process Development Engineers design, implement, manufacture, mine, and other production systems using technical knowledge and expertise in the industry. They use computer modeling software to test technologies and machinery. An individual who is opting career as Process Development Engineer is responsible for developing cost-effective and efficient processes. They also monitor the production process and ensure it functions smoothly and efficiently.

AWS Solution Architect

An AWS Solution Architect is someone who specializes in developing and implementing cloud computing systems. He or she has a good understanding of the various aspects of cloud computing and can confidently deploy and manage their systems. He or she troubleshoots the issues and evaluates the risk from the third party.

Azure Administrator

An Azure Administrator is a professional responsible for implementing, monitoring, and maintaining Azure Solutions. He or she manages cloud infrastructure service instances and various cloud servers as well as sets up public and private cloud systems.

Computer Programmer

Careers in computer programming primarily refer to the systematic act of writing code and moreover include wider computer science areas. The word 'programmer' or 'coder' has entered into practice with the growing number of newly self-taught tech enthusiasts. Computer programming careers involve the use of designs created by software developers and engineers and transforming them into commands that can be implemented by computers. These commands result in regular usage of social media sites, word-processing applications and browsers.

Information Security Manager

Individuals in the information security manager career path involves in overseeing and controlling all aspects of computer security. The IT security manager job description includes planning and carrying out security measures to protect the business data and information from corruption, theft, unauthorised access, and deliberate attack

ITSM Manager

Automation test engineer.

An Automation Test Engineer job involves executing automated test scripts. He or she identifies the project’s problems and troubleshoots them. The role involves documenting the defect using management tools. He or she works with the application team in order to resolve any issues arising during the testing process.

Applications for Admissions are open.

Aakash iACST Scholarship Test 2024

Get up to 90% scholarship on NEET, JEE & Foundation courses

SAT® | CollegeBoard

JEE Main Important Chemistry formulas

As per latest 2024 syllabus. Chemistry formulas, equations, & laws of class 11 & 12th chapters

Resonance Coaching

Enroll in Resonance Coaching for success in JEE/NEET exams

TOEFL ® Registrations 2024

Thinking of Studying Abroad? Think the TOEFL® test. Register now & Save 10% on English Proficiency Tests with Gift Cards

ALLEN JEE Exam Prep

Start your JEE preparation with ALLEN

Everything about Education

Latest updates, Exclusive Content, Webinars and more.

Download Careers360 App's

Regular exam updates, QnA, Predictors, College Applications & E-books now on your Mobile

Cetifications

We Appeared in

Essay on My Role Model

Students are often asked to write an essay on My Role Model in their schools and colleges. And if you’re also looking for the same, we have created 100-word, 250-word, and 500-word essays on the topic.

Let’s take a look…

100 Words Essay on My Role Model

Introduction.

My role models are my parents. They are the pillars of strength and guide me in my life.

Parental Love

Their unconditional love and care make me feel secure. They teach me to love and respect all.

Hard Work and Dedication

My parents work hard to provide for our family. Their dedication inspires me to work hard in my studies.

Values and Morals

They instill in me good values and morals, shaping me into a better person.

Thus, my parents are my role models, guiding me towards a bright future.

250 Words Essay on My Role Model

Parents are the first role models for any child. In my life, my parents have played a pivotal role, shaping my character, values, and aspirations. Their influence is not confined to my early years but continues to guide me in my adult life.

Embodiments of Values

My parents are the embodiment of core values such as honesty, hard work, and compassion. Their consistent demonstration of these values has deeply ingrained them in my character. They’ve taught me that integrity is not situational but a constant trait, and hard work is the key to success.

Inspiration and Guidance

They have always been my guiding light, helping me navigate through life’s complexities. Their life stories, filled with struggles and triumphs, inspire me to face challenges with courage and resilience. They have shown me that setbacks are not the end but stepping stones to success.

Unconditional Love and Support

Their unconditional love and support have been my safety net. They have taught me the importance of empathy and understanding, shaping me into a compassionate individual. Their unwavering faith in me has instilled confidence and encouraged me to pursue my dreams.

My parents’ influence extends beyond their roles as caregivers. They are my role models, guiding me through life’s journey with their wisdom, instilling in me values, and nurturing my dreams. Their impact is profound and enduring, shaping me into the person I am today.

500 Words Essay on My Role Model

In the grand tapestry of life, there are few figures as influential and formative as our parents. They are our first teachers, our primary guides, and the mirrors through which we perceive ourselves and the world around us. As a college student, I’ve come to appreciate the profound impact my parents have had on my life, shaping me into the person I am today. They are my role models, their values and principles serving as a compass guiding my journey.

My Parents: Pillars of Strength and Wisdom

My parents, with their unique blend of strength and wisdom, have always been a beacon of inspiration. Their personal and professional journeys, marked by resilience and determination, have taught me that life is not about avoiding storms but learning how to dance in the rain. They have shown me that success is not measured by wealth or status, but by the positive impact we have on others’ lives.

The Value of Education

As staunch advocates for education, my parents instilled in me a love for learning from an early age. They emphasized that education is not merely about acquiring knowledge or skills, but about nurturing curiosity, developing critical thinking, and fostering a lifelong commitment to personal growth. This perspective has shaped my academic journey, encouraging me to pursue knowledge beyond textbooks and classrooms, and to value intellectual curiosity over rote learning.

Lessons in Empathy and Kindness

My parents’ lessons extend beyond the academic realm. They have demonstrated the importance of empathy and kindness, teaching me that compassion is the cornerstone of human connection. Their acts of generosity, both big and small, have shown me that kindness is not merely an act but a way of life. These lessons have shaped my interactions with others, encouraging me to approach every individual with understanding and respect.

Resilience and Perseverance

Life is not always smooth sailing, and it is during these turbulent times that my parents’ resilience shines through. They have taught me that setbacks and failures are not roadblocks but stepping stones towards success. Their perseverance in the face of adversity has inspired me to embrace challenges with courage and determination, and to view each obstacle as an opportunity for growth.

In conclusion, my parents are my role models, their lives a testament to the power of perseverance, the value of education, and the importance of kindness and empathy. They have shaped my worldview, inspired my aspirations, and guided my actions. As I navigate the complexities of college life and beyond, I am grateful for their unwavering support and guidance. Their footprints serve as a roadmap, guiding me towards a life of purpose, fulfillment, and positive impact.

That’s it! I hope the essay helped you.

If you’re looking for more, here are essays on other interesting topics:

Essay on Homework
Essay on Charity Begins at Home
Essay on My Hometown

Apart from these, you can look at all the essays by clicking here .

Happy studying!

Why My Mom is My Role Model

Table of contents, unwavering strength and resilience, compassion and empathy, leading by example.

*minimum deadline

Cite this Essay

To export a reference to this article please select a referencing style below

Related Essays

Need writing help?

You can always rely on us no matter what type of paper you need

*No hidden charges

100% Unique Essays

Absolutely Confidential

Money Back Guarantee

By clicking “Send Essay”, you agree to our Terms of service and Privacy statement. We will occasionally send you account related emails

You can also get a UNIQUE essay on this or any other topic

Thank you! We’ll contact you as soon as possible.

Open access
Published: 16 May 2023

Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling

Yuan-Peng Zhang 1 , 2 , 3 na1 ,
Xin-Yun Zhang 1 na1 ,
Yu-Ting Cheng 1 na1 ,
Bing Li 4 na1 ,
Xin-Zhi Teng 2 ,
Jiang Zhang 2 ,
Saikit Lam 2 ,
Ta Zhou 2 ,
Zong-Rui Ma 2 ,
Jia-Bao Sheng 2 ,
Victor C. W. Tam 2 ,
Shara W. Y. Lee 2 ,
Hong Ge 4 &
Jing Cai 2 , 3

Military Medical Research volume 10 , Article number: 22 ( 2023 ) Cite this article

4884 Accesses

11 Citations

4 Altmetric

Metrics details

Modern medicine is reliant on various medical imaging technologies for non-invasively observing patients’ anatomy. However, the interpretation of medical images can be highly subjective and dependent on the expertise of clinicians. Moreover, some potentially useful quantitative information in medical images, especially that which is not visible to the naked eye, is often ignored during clinical practice. In contrast, radiomics performs high-throughput feature extraction from medical images, which enables quantitative analysis of medical images and prediction of various clinical endpoints. Studies have reported that radiomics exhibits promising performance in diagnosis and predicting treatment responses and prognosis, demonstrating its potential to be a non-invasive auxiliary tool for personalized medicine. However, radiomics remains in a developmental phase as numerous technical challenges have yet to be solved, especially in feature engineering and statistical modeling. In this review, we introduce the current utility of radiomics by summarizing research on its application in the diagnosis, prognosis, and prediction of treatment responses in patients with cancer. We focus on machine learning approaches, for feature extraction and selection during feature engineering and for imbalanced datasets and multi-modality fusion during statistical modeling. Furthermore, we introduce the stability, reproducibility, and interpretability of features, and the generalizability and interpretability of models. Finally, we offer possible solutions to current challenges in radiomics research.

Cancer is a devastating disease that affects many people worldwide [ 1 ]. Cancerous tumors start as a small cluster of neoplastic cells that may be located within an intricate network of internal tissues and organs, which makes it difficult to diagnose such cancers (e.g., nasopharyngeal carcinoma) in their early stages [ 2 ]. In addition, cancers of the same type and stage may behave remarkably differently in different patients, so it is critical that methods are available to monitor tumor growth, to assist clinicians in prescribing anti-cancer treatment, and to assess treatment responses in individual patients [ 3 ].

In this regard, medical imaging, such as computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), and ultrasonography (US), is indispensable for detecting the presence and monitoring the growth of cancer, and assessing treatment responses. Different imaging modalities capture different properties of internal organs. For instance, CT detects anatomical changes in organs, such as arterial calcification [ 4 ]; MRI visualizes soft-tissue contrast and the musculoskeletal system [ 5 ]; and PET captures functional and metabolic changes in tissues or organs [ 6 ]. Contrast agents are often employed to enhance visualization of the contrast between signal intensities in images of normal and abnormal tissues (such as tumors). Nonetheless, clinical judgement based on unaided visual inspection of images can be resource intensive, is dependent on physicians’ experience, and may fail to detect all of the information within the three-dimensional (3D) volume of a tumor.

Radiomics has recently emerged as a promising solution to these problems, as it involves high-throughput extraction and analysis of high-dimensional quantitative features from multi-modal medical images [ 7 ], which enables it to non-invasively capture intratumoral heterogeneity [ 8 ]. Radiomics-based studies consist of the following six steps: image acquisition, image preprocessing, image segmentation, feature extraction, feature selection, and model construction and evaluation [ 9 ]. The key steps are those involved in feature engineering (i.e., feature extraction and feature selection) and statistical modeling (i.e., model construction and evaluation) and are the current focus of most researchers’ efforts. Moreover, good progress in feature engineering and statistical modeling has been made in recent years. For example, radiomics features are now known to be correlated with tumor diagnosis and prognosis, so researchers have used the minimal redundancy maximal relevance (mRMR) method, the least absolute shrinkage and selection operator (LASSO), and other technologies to select predictive radiomics features. They have also used classifiers such as support-vector machine (SVM) and random forest (RF) to construct radiomics-based models. Numerous studies have also constructed radiomics-based models to, for example, aid in cancer diagnosis, prognosis, and the prediction of treatment responses. These models have demonstrated the possibility of developing risk stratification and personalized treatment for patients, which could lead to the realization of precision medicine. However, despite this progress in radiomics, several key problems remain to be solved.

In this review, we summarize recent literature on applications of radiomics for the investigation of tumors, with a particular focus on feature engineering and statistical modeling methods. We also review aspects that may influence model performance, such as feature stability and model generalizability; highlight the problems that remain to be comprehensively solved (such as imbalanced datasets and multi-modality fusion); and make recommendations to the community for future research and development.

Clinical applications based on radiomics

Radiomics-based cancer diagnosis.

Traditional medical imaging plays an important role in the diagnosis of cancer, but misdiagnoses and missed diagnoses nevertheless occur. These are major problems, as they prevent early diagnoses and thus timely clinical intervention, thereby decreasing cancer patients’ survival rates and cure rates [ 10 , 11 ]. This problem can potentially be addressed by the augmentation of traditional medical imaging with radiomics, which can capture phenotypic information of tumors [ 12 ] and has shown promise in differentiating benign and malignant tumors and predicting treatment responses. Radiomics analysis relies on artificial intelligence (AI) algorithms, which can improve the accuracy (Acc) of predictive models used for the diagnosis and evaluation of treatment responses. In particular, radiomics applies feature engineering to detect intra-tumoral properties in medical images that are typically undetected during visual inspection of such images by physicians. In the following, we provide a comprehensive overview of the utilization of radiomics-based methods for cancer diagnosis, with a particular focus on three distinct perspectives: tumor grading, tumor staging, and the classification of malignant vs. benign tumors.

Radiomics-based tumor grading

Reliable pre-surgical radiomics-based evaluation of tumor grading can help to formulate treatment plans for patients and can also reduce the recurrence rate and incidence of adverse effects. As shown in Table 1 [ 13 , 14 , 15 , 16 ], radiomics analysis has been explored for tumor grading (i.e., describing the magnitude of tumor atypia) in various cancers like head and neck cancers (HNCs) and lung cancers, as it is an efficient non-invasive method for pathological examination. Specifically, Wu et al. [ 13 ] constructed a radiomics signature with kernel principal component analysis (KPCA), a RF classifier, and a variance-threshold, which they used to develop a radiomics model. They then compared the performance of this model against that of a clinical model and a combined clinical-radiomics model in the analysis of CT images for the grading of head and neck squamous cell carcinoma (HNSCC) tumors. They found that the combined model outperformed the other two models, as their respective areas under the receiver operating characteristic curve (AUCs) were 0.97, 0.96 and 0.63. Mukherjee et al. [ 14 ] used CT-based radiomics features to develop models for HNSCC tumor grading, predicting perineural invasion, and lymphovascular invasion, and these models’ AUCs were 0.66, 0.70, and 0.65, respectively. Although these performances suggest that these models are unsuitable for clinical adoption, they may be sufficient to demonstrate the potential of CT-based radiomics features for predicting histopathologic characteristics.

Radiomics-based tumor staging

Some studies have focused on developing radiomics-based models for tumor staging as shown in Table 2 [ 17 , 18 , 19 , 20 ], which classifies the severity of a malignancy according to the size of the primary tumor and the extent of its spread throughout the body. Traditionally, tumor staging information is obtained by performing imaging examinations and pathological biopsies; in contrast, radiomics can be applied for preoperative tumor staging and thus may be more effective. For example, we included three radiomics studies [ 17 , 18 , 19 ] that addressed preoperative tumor staging of different cancers. Ren et al. [ 17 ] preoperatively distinguished I–II and III–IV stage HNSCC tumors by analyzing the radiomics features of T2-weighted (T2W) and contrast-enhanced T1-weighted (ceT1W) MR images. They found that the radiomics signatures based on the ceT1W images (AUC = 0.853) best discriminated between stage I–II and stage III–IV HNSCC tumors, followed by models based on T2W and ceT1W combined images (AUC = 0.849). In the study by Gao et al. [ 20 ], a radiomics model based on 30 US features was constructed to evaluate tumor staging. The study employed ten classifiers and observed that the Naive Bayes model attained an AUC of 0.84 in the validation cohort. Furthermore, tumor staging can be performed using CT imaging, MR imaging, and US imaging. It is notable that CT imaging is a less time-consuming and less costly option compared to MR imaging; however, MR imaging is superior in distinguishing soft tissue changes from cartilage abnormalities. Notably, US imaging is less expensive than both CT and MR imaging, but its resolution is inferior to that of CT imaging. Therefore, an appropriate imaging mode should be selected according to the research objectives of a given situation.

Radiomics-based classification of malignant vs. benign tumors

Pathological examination is the gold-standard approach for the diagnosis of benign and malignant tumors. However, it is an invasive approach, and radiomics can serve as a non-invasive alternative. Table 3 [ 21 , 22 , 23 ] summarizes the application of radiomics-based classification of benign and malignant tumors. Ho et al. [ 21 ] identified 89 features in MR images that can be used to discriminate between benign and malignant lymph nodes. The most discriminating of these 89 features is “Original_glcm_DifferenceAverage”, which measures the relationship between occurrences of pairs with similar intensity values and occurrences of pairs with different intensity values. In order to identify benign/malignant liver tumors, Yin et al. [ 22 ] developed a CT-based radiomics model, which achieved an average AUC of 0.87.

The above-described studies have accumulated evidence that radiomics analysis has the potential to serve as a non-invasive diagnostic tool prior to the treatment of cancer. In particular, radiomics analysis can increase the Acc and the objectivity of tumor staging and grading, thereby serving as an auxiliary decision tool for personalizing treatment.

Radiomics-based cancer prognosis prediction

Radiomics may be a better predictor of treatment response, survival, tumor recurrence, or metastasis than other conventional methods [ 24 , 25 , 26 , 27 ]. Therefore, radiomics could assist clinicians to accurately stratify the negative prognosis risks of patients before treatment, select appropriate treatment plans, support the development of personalized medicine, and thereby substantially improve cancer prognosis. Here, we review the applications of radiomics-based prognosis prediction from the aspects of survival, local recurrence, and metastasis.

Radiomics-based survival prediction

Table 4 [ 15 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 ] summarizes recent radiomics-based survival studies that have investigated aspects such as death prognosis, all-cause mortality progression-free survival (PFS), and disease-free survival (DFS). Fh et al. [ 28 ] developed deep learning models based on radiomics features from planning target volumes (PTVs) and gross tumor volumes (GTVs) to simultaneously predict patients’ death and tumor recurrence. Their model based on GTV radiomics features predicted death and tumor recurrence with Accs of 85.9% and 72.4%, respectively, and overall AUCs of 0.947 and 0.956, respectively, whereas their model based on PTV features predicted these two endpoints with Accs of 77.7% and 74.3%, respectively, and overall AUCs of 0.934 and 0.932, respectively. A 2020 study [ 30 ] predicted the PFS time and overall survival (OS) time of oropharyngeal squamous cell carcinoma (OPSCC) by using the American Joint Committee on Cancer (AJCC) staging model [which the AJCC built based on their staging scheme for human papillomavirus (HPV)-related and -unrelated OPSCC] as the baseline model and constructing a PET/CT-based radiomics model. This study found that the average concordance index ± standard deviation of its radiomics model was generally higher than that of the AJCC model in both HPV-related and -unrelated cohorts, which proved that PET/CT radiomics features can add prognostic value beyond that offered by the AJCC staging scheme. Studies [ 31 , 32 , 33 , 34 , 35 ] have also developed radiomics models to predict aspects such as DFS, OS, long-term survival, and recurrence-free survival, further underscoring the utility of radiomics features for predicting cancer survival.

Other studies have focused on determining whether associations exist between selected radiomics features and specific outcomes of patients following chemoradiotherapy. Selecting such features before modeling helps to reduce redundancy and improve the predictive performance of models. Parmar et al. [ 15 ] investigated the prognostic features of lung cancer and HNC by applying consensus clustering to generate 11 and 13 radiomics feature clusters of lung cancer and HNC, respectively, and then located the clusters that were highly correlated with cancer prognosis. Subsequently, they constructed models and evaluated their prognostic performance, which revealed that the models generated inconsistent prognoses for the two diseases because the clusters were site specific.

Radiomics-based recurrence prediction

Folkert et al. [ 29 ] constructed a multivariable predictive model of tumor-related endpoints (all-cause mortality, local failure and distant metastasis) by integrating clinical parameters and fluorodeoxyglucose (FDG)-PET-based radiomics features; this model revealed that metabolic tumor volume (MTV) was correlated with all three endpoints. However, in an independent cohort validation, the multivariable model with local failure (local recurrence) as the endpoint had the highest AUC [0.73 ( P = 0.026)] and was thus the most predictive; in contrast, the models with all-cause mortality and distant metastasis as endpoints, respectively, had lower AUCs [0.65 ( P = 0.004) and 0.66 ( P = 0.015), respectively] and were moderately predictive. There are also radiomics studies about cancer recurrence, which are summarized together with the above study in Table 5 [ 28 , 29 , 36 , 37 , 38 ].

Radiomics-based metastasis prediction

Regarding distant metastasis of cancer, one study [ 39 ] investigated a distant metastasis model based on MRI features, extracting a total of 2803 radiomics features from the MR images of 176 patients. They then screened these features to yield seven features that were used with a logistic regression algorithm to form a distant metastasis model that was superior to a clinical factor-based model. Subsequently, they constructed a nomogram that can help to determine the risk of metastasis for a patient and develop personalized treatment plans.

Lymph node metastasis (LNM) is traditionally adopted to guide decision-making on prescribed treatments for cancer [ 40 ]. The occurrence of LNM usually indicates a poor prognosis, so it is essential to count in LNM when treating tumors. Wang et al. [ 34 ] extracted radiomics features from T2W MR images of primary tumors with and without 3D peritumoral extensions (3, 5, 10, and 15 mm, respectively) and incorporated these features with clinicopathological features. They then constructed combined models using an SVM classifier and found that with the test set, the combined model based on 10‐mm peritumoral extensions achieved a higher AUC (0.872) in predicting LNM than the combined models based on no peritumoral extensions (AUC = 0.720), 3-mm peritumoral extensions (AUC = 0.787), 5-mm peritumoral extensions (AUC = 0.793), and 15-mm peritumoral extensions (AUC = 0.808). A study published in 2020 [ 41 ] analyzed the relationship between radiomics features extracted from US images and LNM of thyroid cancer. This showed that only an irregular shape and microcalcification were effective predictors of LNM, with AUCs of 0.591 ( P = 0.059) and 0.629 ( P = 0.007), respectively. Table 6 [ 29 , 34 , 35 , 39 , 41 , 42 , 43 ] shows the main content of recent radiomics studies of cancer metastasis.

Radiomics-based cancer treatment responses prediction

As treatment responses are closely related to OS, seven studies [ 44 , 45 , 46 , 47 , 48 , 49 , 50 ] have investigated the application of radiomics for constructing models to predict treatment responses to radical radiotherapy, chemotherapy or chemoradiotherapy. For example, one of these studies [ 44 ] used quantitative US (QUS) delta-radiomics to monitor the response of HNC to radical radiotherapy. K-nearest neighbors (KNN) and Naive Bayes algorithms were both used to construct single-, two-, and three-feature models. The results showed that the single-feature Naive Bayes model had the highest Acc in predicting responses after 3 months of treatment: its Acc based on the QUS characteristics at 24 h after chemoradiotherapy was 80%, and increased to 85% when the QUS characteristics obtained at the fourth week were included.

In addition, a dual-center retrospective study [ 47 ] was performed to extract radiomics features from the MR images of 221 patients before induction chemotherapy (IC) and 96 patients after IC, and then calculate the delta-radiomics feature values (by subtracting the feature values from MR images obtained after IC from those obtained before IC) and process them. Subsequently, a pre-treatment MRI radiomics model and a delta-radiomics model were generated and trained using pre-treatment MRI radiomics features and delta-radiomics features, respectively, to predict the tumor retraction response to IC plus concurrent chemoradiotherapy. The trained models were then applied to an external validation set and afforded AUCs of 0.983 and 0.818, respectively, demonstrating their potential utility as references for devising cancer-treatment plans. Table 7 [ 44 , 45 , 46 , 47 , 48 , 49 , 50 ] summarizes recent radiomics-based treatment response studies.

Radiomics-based cancer treatment complications prediction

Radiation toxicity is an important consideration for treatment optimization, and its accurate prediction allows enhanced personalization of treatment plans. Many studies have investigated adverse effects of radiotherapy and chemotherapy on tissues. For instance, radiotherapy for nasopharyngeal carcinoma may cause cervical spine osteoradionecrosis [ 51 ], which is difficult to distinguish from bone metastasis by visual inspection of images. However, radiomics was demonstrated to have great potential for the accurate detection of cervical spine osteoradionecrosis. For example, Zhong et al. [ 52 ] used a LASSO logistic regression algorithm based on tenfold cross-validation of the minimum criteria to select eight relevant features, which they used to develop a radiomics nomogram that can distinguish osteoradionecrosis from cancer bone metastasis. The AUC of this nomogram reached 0.72 in the validation cohorts. Xerostomia is the most common side-effect of radiotherapy for HNC, and its prediction based on radiomics features has been extensively investigated. In a study published in 2018 [ 53 ], the lesion regions of interest (ROIs) on MR images of parotid glands were delineated using the target contour of CT images. Then, a reference model that predicts xerostomia based only on the parotid gland radiotherapy dose and patient-reported xerostomia at the start of radiotherapy was fitted. The reference model had an AUC in the external validation of 0.65, which was increased to 0.83 by the addition of quantified MRI features, thereby improving its ability to predict the occurrence of xerostomia.

Most of the radiomics studies conducted to date have been retrospective, which can lead to overestimation of the performance of radiomics models. In contrast, applying these models to prospective studies requires a sufficiently large training cohort and strong performance. Prospective studies also involve recruiting participants in advance without relying on existing patients’ data, and conducting long-term follow-ups to evaluate the results predicted by the model. While the results of these studies cannot be communicated to clinicians and participants, they provide a more reliable estimate of model performance. The excellent results obtained in these studies demonstrate the high generalizability and robustness of radiomics, making it a viable tool for clinical applications. The six prospective studies we reviewed [ 35 , 38 , 43 , 48 , 49 , 50 ] showed acceptable performance of the radiomics model, with AUCs ranging from 0.688 to 0.871. Although the lowest AUC value of 0.688 [ 35 ] suggests that the model has limited classification performance, it still demonstrates predictive power. The study with the highest AUC value of 0.871 [ 48 ] also had an acceptable number of participants in the prospective cohort.

AI-driven radiomics studies

The workflow and challenges of radiomics-based studies are illustrated in Fig. 1 , based on which we discuss feature engineering and statistical modeling. Feature engineering focuses on the reproducibility, and interpretability of features, whereas statistical modeling focuses on the generalizability of a given model, imbalanced data classification, multi-modality fusion of the model, and interpretability of the model. In addition, feature reproducibility is one of the factors influencing model generalizability. Thus, an improvement in feature reproducibility can enhance model generalizability, which means that the model can be popularized better in clinical practice.

Workflow and challenges of radiomics-based studies. The workflow of radiomics involves several key stages including image acquisition, image reconstruction and preprocessing, image segmentation, feature extraction and selection, model construction, and internal and external validation. However, there are still several challenges that need to be addressed in this workflow such as ensuring feature stability and reproducibility, improving model generalizability and interpretability, addressing imbalanced data classification, and improving multi-modality fusion in statistical modeling. It is worth noting that feature engineering and statistical modeling are two important components of radiomics

In this section, we review studies that have offered insights into and suggestions on how to solve feature engineering and statistical modeling challenges in radiomics.

Feature engineering

Feature extraction

Feature extraction involves comprehensively quantifying the tumor phenotypes based on high-throughput features that are hypothetically associated with the tumor microenvironment. Specifically, after completion of some steps of the radiomics workflow defined above, such as image acquisition, preprocessing, and segmentation, radiomics features associated with a given research purpose are extracted from within two-dimensional (2D) or 3D ROIs in images. The selection of ROIs is diverse. The majority of radiomics studies discussed non-metastatic carcinoma (M0) population, so the primary tumor generally is the ROI for feature extraction. On the other hand, there are a few radiomics articles that studied the metastatic population (M1) [ 54 ], therefore it is possible for them to select the metastasis site as the ROI for feature extraction. In addition, the intra-tumoral region and the peritumoral region can also be ROI for feature extraction, and the features of the peritumor region may show better prognostic performance [ 55 ]. Therefore, the selection of ROIs primarily depends on the study population.

These are generally distinctive types of radiomics features of ROIs, such as shape, first-order, and texture features (Fig. 2 ). Shape features define the shape of an ROI, such as its sphericity, volume, and surface area. First order features describe the properties of histogram, regardless of the spatial relationship [ 56 ]. such as mean value, median value and entropy. Texture features describe the properties of secondary matrix, such as gray level co-occurrence matrix features [ 57 ], gray level run-length matrix features [ 58 ], gray level dependence matrix features [ 59 ], gray level size zone matrix features, and neighboring gray tone difference matrix features. It follows that textural features can quantify the spatial relationships between voxels. Furthermore, radiomics features can be extracted from original images, log-sigma-filtered images, and wavelet-filtered images, with those extracted from the latter two types of images generally referred to as higher-order features.

Categories of radiomics features. Radiomics features including shape features, first order features and texture features. a Shape features describe the shape of ROI such as sphericity and volume. b First order features, such as mean value, median value and entropy, are calculated based on the histogram to describe the distribution of individual voxels values, regardless of the spatial relationship. c Texture features quantify the spatial relationship between voxels, which obtained from various matrix types, such as gray level co-occurrence matrix features, gray level run length matrix features, gray level dependence matrix features, gray level size zone matrix features, and neighbouring gray-tone difference matrix features

Generally, 2D features are extracted from single-layer images containing the most typical or largest cross-section of a lesion, whereas 3D features are calculated from an entire ROI volume. Thus, compared with 2D features, 3D features contain more tumor information but may also contain more interference due to, for example, noise and variations in slice thickness, and are less easy to obtain, slower to calculate, and more labor-intensive to generate (due to multiple-layer contouring being required) [ 60 ]. As a result, whether to use 2D or 3D image features remains a topic of debate. Wan et al. [ 61 ] evaluated the diagnostic performance of 2D and 3D radiomics features based on MR images of solitary pulmonary lesions (SPL). They found that the latter features performed better than the former features (AUCs: 0.824 vs. 0.740) and that combined features did not show better performance than either type of features alone (AUC = 0.813). Xu et al. [ 62 ] found that 3D radiomics features showed better predictive performance than 2D radiomics features in a study of multi-organ cancer, as unlike the latter, the former was significantly correlated with total lesion glycolysis, tumor volume, and staging [ 63 ]. However, Shen et al. [ 60 ] demonstrated that compared with 3D radiomics features, 2D radiomics features of CT images of non-small cell lung cancer (NSCLC) performed slightly better, and Zhu et al. [ 64 ] reached the same conclusion. Both studies have attributed these performance discrepancies between 2 and 3D radiomics features to the inconsistent resolution of CT images. However, there is no conclusive evidence that 2D features are superior to 3D features.

Researchers have also been searching for new quantitative imaging features to enrich radiomics investigations. Beichel et al. [ 65 ] evaluated the ability of 17 features of PET images, such as standardized added metabolic activity and rim average (RA; the mean of uptake in a two-voxel-wide rim region around an ROI), to predict the DFS of HNC patients. They found that RA may help differentiate between true- and false-positive recurrences of HNC. Buizza et al. [ 66 ] devised a new set of PET/CT image radiomics features (longitudinal patterns) to capture changes in the intensity at various distances from the border of a tumor. This set of radiomics features is superior to traditional radiomics features, as the latter are extracted from a defined ROI, meaning that useful information elsewhere in a tumor is often undetected. Moreover, there are two studies [ 44 , 47 ] that developed a new form of radiomics features. That is, the researchers extracted features from images before and after treatment and then subtracted the post-treatment features from the pre-treatment features to obtain the radiomics features that were used to predict treatment response. These features can help to quantify peritumoral information that is complementary to intratumoral radiomics features. However, these quantitative imaging features are modality-dependent and have not been standardized, so they cannot yet be used as conventional radiomics features.

At present, radiomics studies typically extract large numbers of features from images using commercial or open-source software or software package, such as PyRadiomics [ 67 ], 3D Slicer [ 68 ], and Imaging Biomarker Explorer [ 69 ]. These radiomics features can be calculated by corresponding formulas. The Image Biomarker Standardization Initiative [ 70 ] was established to standardize the extraction of image features and thereby ensure the repeatability of feature extraction across different platforms.

Feature selection

Typically, hundreds of radiomics features are extracted as modeling candidates, and if all of these were used to construct a model, it would have excessive feature dimensions and be too complex, meaning that it would over-fit data and thus have low generalizability. Furthermore, most extracted radiomics features are highly correlated with each other, so reduction and feature selection must be performed before modeling. This is achieved using radiomics feature-selection methods, which select the most relevant features and remove the redundant features from a large number of features. Feature reproducibility should be considered during feature selection, as the aim of the latter is to obtain the optimal feature subset or feature representation that has the maximum correlation with the endpoints and the minimum correlation with other features [ 71 ].

Feature selection methods comprise filtering, embedded, and wrapper methods (Fig. 3 ). Filtering methods rank features according to the repeatability and their relevance to the endpoints. Then, the top ranked features or those that are above a specified threshold value are selected or excluded. Independent features are filtered by using the Pearson correlation method to exclude features with, for example, correlation coefficient > 0.75 (or some other pre-determined thresholds of correlation coefficient). Thus, in studies (e.g., [ 72 ]) that have used the Pearson correlation method to assess the correlation between tumor volume and radiomics feature values, highly volume-correlated features that meet a Pearson’s correlation threshold have been removed. Another filtering method is mRMR method [ 73 ], which aims to identify the best subset of features, maximize the relevance between subset and target variables, and minimize the redundancy between features based on mutual information. Hu et al. [ 74 ] used the mRMR method for dimensionality reduction in a radiomics study of nasopharyngeal carcinoma. Other filtering methods that have been used are Relief [ 75 ], Student’s t -test [ 76 ], and Chi-square test [ 77 ]. In addition, Parmar et al. [ 78 ] examined 14 filtering methods and found that features selected using the Wilcoxon test showed high stability (stability = 0.84 ± 0.05) in their training cohort. Wrapper methods employ model performance as a criterion to judge the quality of features or a feature subset; that is, they gradually retain or remove several features and finally select the feature subset that enables a given model to achieve optimal performance. For instance, recursive feature elimination (RFE) is widely used in radiomics: it generates a subset of features, iteratively constructs a model from the current feature subset, obtains the degree of importance of each feature, removes unimportant features, and retains the features with the best performance [ 79 ]. Yu et al. [ 80 ] adopted RFE for feature selection in their multiphasic CT-based radiomics analysis to differentiate benign and malignant parotid tumors, and used multiple methods for feature dimensionality reduction. In contrast to filtering methods and wrapper methods, embedded methods perform feature selection and model training simultaneously. First, a classifier obtains the weight coefficient of each feature after training, and then these coefficients are evaluated by a specific model to select the best feature, i.e., the feature is directly selected by the model. LASSO [ 81 ] is a commonly used embedded method that applies regularization to remove redundant features and retains the most relevant features. However, LASSO tends to ignore the pairwise correlations of features [ 82 ], so it must be combined with other feature redundancy elimination methods to enhance model reliability. In most radiomics studies, feature selection has been conducted via multiple steps using a combination of methods focused on different feature characteristics. For example, in a study of nasopharyngeal carcinoma [ 83 ], intraclass correlation coefficients (ICC) were first used to evaluate inter- and intra-observer agreement, and features with high reproducibility were selected. Then, the Wilcoxon rank sum test was used to select the radiomics features that statistically differed between regions of lymphatic infiltration and regions of non-lymphatic infiltration. Finally, LASSO was used to select the most relevant and independent features from a training set.

Feature selection. Feature selection methods including filtering, wrapper, and embedded. a The filtering methods rank the features according to a certain characteristic or correlation, and specify a threshold value or directly select the top ranked features. b The embedded method adopts the way that the feature is directly selected by the model. The model obtains the weight coefficient of each feature after trainings and selects the best feature according to the coefficient. c The wrapper methods take model performance as a criterion to judge the quality of features or feature subsets, and gradually retain or remove several features

The feature selection methods described above are supervised methods. However, researchers have also employed unsupervised methods, such as principal component analysis (PCA) [ 84 ], and t -distributed stochastic neighbor embedding [ 85 ]. However, most radiomics studies have implemented supervised methods, so unsupervised methods are not reviewed here.

Feature reproducibility

Reproducibility is a measure of the variability of repeated measurements of the same or similar quantitative imaging biomarkers in a real clinical environment and is affected by external factors that cannot be strictly controlled, such as operators, measurement systems, and measurement equipment [ 86 , 87 ]. Thus, reproducibility represents stability, so radiomics studies must ensure that the radiomics features they use have high reproducibility, such that their models generate similar classification results in different clinical environments. The values of features are affected by all of the steps prior to radiomics analysis, including image acquisition and preprocessing [ 88 ]. Therefore, during research, appropriate treatment should be performed as far as possible from the source of variation to obtain stable features with high reproducibility. Based on the radiomics workflow, we discuss the stability of features in terms of image acquisition, image preprocessing, characterization or segmentation of tumor areas of interest, and feature selection (as shown in Fig. 4 ).

Influencing factors of feature reproducibility. The alteration of image acquisition details among the five influential factors can significantly affect the extracted features, resulting in varying outcomes. The stability of radiomics features is highly influenced by image reconstruction techniques. The commonly used filtered back-projection and iterative methods in radiomics research tend to decrease the stability of radiomics features. The test–retest strategy can be an effective tool in reducing the variability caused by image acquisition and reconstruction. The choice of ROI segmentation methods such as manual, semi-automatic, and automatic, and the size of ROI can contribute to different levels of feature reproducibility. Moreover, inter-machine reproducibility has a substantial influence on the degree of feature variation. ROI region of interest

Features are extremely sensitive to changes in acquisition details, even two images of the same tissue may yield different results due to differences in acquisition details [ 89 ]. This sensitivity usually affects the generalization performance of a final model. Balagurunathan et al. [ 90 ] conducted a test–retest study of lung CT images and found that the concordance correlation coefficient (CCC) ≥ 0.9 of radiomics features was only 30.14%. As it is unlikely that a tumor changes within a short period, these unstable radiomics features might have been due to the different postures of patients during rescanning. Midya et al. [ 91 ] found that image acquisition parameters (i.e., tube current and noise index) and reconstruction techniques strongly affected the reproducibility of CT-based radiomics features. There is inevitable noise interference in image acquisition, and Tu et al. [ 92 ] found that in the presence of the quantum noise inherent in CT images, the “ShortRunHighGrayLevelEmpha”, “ShortRunLowGrayLevelEmpha”, “LowGrayLevelRunEmpha” and “LongRunLowGrayLevelEmpha” features were the most stable, whereas the cluster shadow and maximum probability features were the most unstable. Image noise can also be reduced by increasing the tube current, as this increases the reproducibility of radiomics features [ 91 ].

Image reconstruction has a strong influence on the stability of radiomics features. For example, the filtered back-projection and iterative methods commonly used in radiomics research decrease the stability of radiomics features. Abundant noise is generated by filtered back-projection methods, but this can be removed (without changing noise texture) by reconstruction using deep-learning neural networks [ 93 ]. Yan et al. [ 94 ] explored the influence of reconstruction settings on textural parameters and found that they were influenced more by grid size than by the number of iterations or the full width at half maximum. Galavis et al. [ 95 ] determined that in different acquisition modes and using different reconstruction parameters, entropy-first order, energy, maximal correlation coefficient, and low-gray level run emphasis parameters exhibited small variations, which means that they have good reproducibility and can be considered good candidates for automatic tumor segmentation. Prayer et al. [ 96 ] explored the reproducibility of CT radiomics features of fibrosing interstitial lung disease (fILD) and found that slice thickness had a more significant impact than reconstruction kernels on the reproducibility of features between and within scanners. Compared with thin slices, thick slices are more appropriate for measuring tumor volume and volume changes [ 97 ], as thin slices increase noise levels, which can obscure texture features. However, thick slices reduce noise levels, but it also blurs the image.

Table 8 [ 95 , 96 , 98 , 99 , 100 , 101 , 102 ] lists studies that have investigated the reproducibility of radiomics features from the perspectives of scanner models or reconstructed environments and have focused on the identification of reproducible features. However, the reproducibility of these radiomics features cannot be directly compared or used.

Voxel-size resampling is a preprocessing step in image acquisition and reconstruction. It is important in CT, where voxel sizes affect a considerable proportion of radiomics features [ 103 ]. Voxel-size resampling can be accomplished using various interpolation algorithms, but these may use different resampling voxel sizes and box widths and thus may modify radiomics feature estimates in different ways [ 104 ]. Therefore, image interpolation should be performed at the same voxel size as much as possible. In addition, resampling may not be sufficient for some texture features. Thus, Shafiq-Ul-Hassan et al. [ 105 ] enhanced feature robustness through voxel size normalization, and Jensen et al. [ 106 ] corrected variability across different volumes of interest by converting CT images into parametric maps with a fixed voxel size. Bologna et al. [ 107 ] examined MR image radiomics and found that image preprocessing methods (Z-score normalization, resampling, Gaussian filtering, and bias field correction) significantly increased the robustness of radiomics features to different sources of variability (time of repetition and echo, voxel size, random noise, and intensity non-uniformity). However, Li et al. [ 108 ] found that image resampling, intensity normalization, and N4 bias field correction did not significantly affect the reproducibility of radiomics features, but the ComBat harmonization method removed most scanner effects and improved the reproducibility of features.

ComBat harmonization is a normalization technique that is widely used in radiomics, as radiomics features are easily affected by differences in acquisition equipment and reconstruction parameters, especially in multicenter studies. ComBat harmonization reduces these differences to lessen their impact on features, which enhances feature reproducibility [ 109 ]. For example, ComBat harmonization effectively eliminated the differences in MR radiomics feature values caused by heterogeneity of multicenter techniques, thus preventing reproducibility being affected [ 110 ]. As ComBat harmonization is a data-driven approach, it can be applied directly to extracted image features (i.e., without the need for retrieval of images), but it is recommended to be applied only after careful examination of the distribution of eigenvalues at the sites to be aggregated [ 111 ]. Crucially, ComBat harmonization eliminates the center effect while preserving some biological information associated with radiomics features [ 56 , 112 ]. ComBat harmonization techniques have also been used in PET or CT radiomics studies. Figure 5 shows the basic workflow of ComBat harmonization.

Workflow of ComBat harmonization. The multicenter data is obtained, and then the features in the images are extracted. The differences of the feature values are harmonized to obtain the normalized features

Apart from applying ComBat harmonization, applying the same and appropriate reconstruction methods as much as possible and conducting test–retest studies are other ways to reduce variability in image acquisition and reconstruction. Manual segmentation remains the first choice in radiomics research; however, it may be replaced gradually by automatic and semi-automatic segmentation, given the continuous improvements in the Acc of the latter methods. Compared with manual segmentation, automatic segmentation reduces inter-observer variability, leads to better reproducibility and robustness, and is faster [ 113 , 114 ]. The reproducibility of manual segmentation can be increased by using multiple experts to perform segmentation. Usually, segmentation was performed by two or three experienced experts and then evaluated by comparing ICC. In addition, contouring protocols or guidelines can be used to reduce inter-expert variation in target volume delineation [ 115 ]. Gitto et al. [ 116 ] used a time-saving method based on geometric transformations of an ROI, which can simulate multiple manual delineations, to evaluate PET/CT radiomics feature stability. They found that over 76% of radiomics features were stable to ROI transitions. However, Jensen et al. [ 88 ] emphasized that ROI size must be considered in radiomics, as MR-derived features are more robust than CT-derived features to changes in ROI size. Denzler et al. [ 117 ] found by examining phantoms that a larger ROI corresponded to a higher percentage of intensity feature stability and suggested that non-contrast-enhanced CT lung images exhibit tissue- and disease-specific radiomics feature stabilities.

New feature-selection methods have also been developed to improve stability. Bologna et al. [ 118 ] developed a method that can be used to assess feature stability and perform preliminary feature selection based on a single acquisition and an ROI. Yan et al. [ 119 ] devised a novel method, named stability selection supervised PCA, that identifies stable features from radiomics big data and applies dimension reduction to achieve right-censored survival outcomes. Lam et al. [ 120 ] constructed a selection algorithm that determines optimal feature selection combinations. This algorithm also ensures selected features to have good AUCs and stability scores. Feature output stability is assessed on the basis of 10 iterations, and the stability scores are quantified by a frequency-based criterion. The retained test dataset for each iteration is used to evaluate the AUC. The product of the two scores for each feature selection combination is determined, and the combination with the highest score product represents the optimal feature selection combination. Flouris et al. [ 89 ] developed a CT simulator that reconstructs images under different noise levels using different reconstruction algorithms, which may have novel applications in automatic processing of multicenter datasets.

Compared with features with low reproducibility, features with high reproducibility exhibit greater resilience to environmental changes. Radiomics studies usually quantify the reproducibility or stability of features using several types of indicators, such as ICC, coefficient of variation (COV), or CCC. These indicators are commonly used to measure the inter-observer stability of radiomics features but may also be used to perform dimension reduction for feature selection (as mentioned in the subsection: Feature engineering). ICC and CCC are the most commonly used indicators, for which threshold values are typically assigned to allow the relative stability of features to be determined. However, there is no consensus on the threshold values for these two indicators, and ICC values obtained from a test–retest analysis cannot be directly compared with those obtained from an inter-observer analysis [ 121 ]. Furthermore, the COV index is often used to provide information on the variability of a feature measurement unit [ 121 ].

Feature interpretability

Radiomics mining is entirely data-driven and interprets imaging data quantitatively rather than qualitatively, meaning that it can obtain much information that is difficult to obtain visually. Radiomics features can also be well defined at a mathematical level, which endows them with a certain level of interpretability. However, there is a lack of interpretability of radiomics features at the biological level, which may limit the application and development of radiomics in medicine. In recent years, there has been an increase in radiomics studies based on deep learning, and deep learning-based radiomics models can outperform conventional radiomics models [ 122 , 123 ]. However, deep learning-based radiomics is a “black box”, as deep features do not have accurate formulations and definitions and thus cannot be conceptualized. Moreover, although deep features can be explained at the feature level by methods that link them to traditional radiomics features and semantic features [ 124 ], the interpretability of deep features remains low. In this part, we focus on the interpretability of traditional radiomics features.

Although some studies (such as those described in the clinical applications based on radiomics section) have demonstrated the predictive and diagnostic power of radiomics in applications related to cancer, the interpretability of radiomics features is limited and does not meet the needs of clinical experts. Therefore, associated biological backgrounds must be supplied with radiomics features to increase their interpretability. As shown in Fig. 6 , this is generally achieved via three approaches: by determining the biological significance of features, by quantifying tumor heterogeneity, and by developing methods to improve feature interpretability.

Three pathways of improving feature interpretability. The interpretability of radiomics features at the biological level can be enhanced by providing information about cancer phenotypes and tumor microenvironment, quantifying tumor heterogeneity, and a number of methods of improving interpretability

Radiomics features provide information about a cancer’s phenotype and a tumor’s microenvironment that is distinct from and complementary to other data, such as clinical and genomic data [ 8 ]. Aerts et al. [ 125 ] used gene set enrichment analysis to compare radiomics features with gene expression profiles and found that the features were significantly correlated with different biologic gene sets. Moreover, Rifi et al. [ 126 ] found that there were significant differences in the radiomics features between different cell lines, suggesting that features can be grouped according to their behaviors. As radiomics features themselves contain information relevant to a given research objective, and this information may be related to gene expression, identification of this information can offer biological perspectives that are not offered by traditional imaging.

Identification of tumor heterogeneity during tumor treatment can help evaluate the effectiveness of treatment and thus guide treatment planning, and also reveals the individuation of a patient’s tumor. Therefore, exploring the association between tumor heterogeneity and texture features can enhance the interpretability of radiomics features. Some first-order histogram features such as kurtosis, skewness, percentiles, and their respective changes are reliable quantitative proxies of tumor heterogeneity and more directly correlated than other features with potential physiological structural changes that occur during the progression of both treated and untreated tumors [ 127 ]. However, one disadvantage of histogram-based estimation of tumor heterogeneity is that it ignores the spatial structure of a tumor [ 128 ]. Wang et al. [ 129 ] divided patients into groups according to their radiomics scores and found that the tumor immunity and tumor microenvironment of the high- and low-scoring groups were different, indicating that radiomics could reflect the heterogeneity of tumors. Similarly, other researchers found that CT features based on the fourier transform are potentially useful for quantifying tumor heterogeneity in lung cancer patients and that radiomics features associated with tumor heterogeneity are correlated with OS [ 130 ]. Moreover, texture features describe the distribution pattern of voxels and can be used to quantify intra-tumor heterogeneity [ 131 ]. Thus, radiomics features can be employed to quantify tumor heterogeneity, and the correlation features of histograms can enhance understanding of tumor heterogeneity.

Several methods have been developed to improve the interpretability of radiomics analyses. Vuong et al. [ 132 ] devised a method for creating radiomics feature-activation maps that allows identification of spatial-anatomical locations responsible for signature activation based on local radiomics. Kuthuru et al. [ 133 ] adopted a dictionary learning approach to derive visually interpretable imaging features. In addition, Luo et al. [ 134 ] proposed an approach which enables exploration of hierarchical relationships between biophysical features based on a Bayesian network.

In addition to the above-described three approaches, the literature suggests another approach that can be used to improve the interpretability of radiomics features from a biological or clinical-physiological perspective. This approach is based on the fact that the semantic features of clinical reports may be more explanatory than traditional radiomics features. Therefore, quantification of explanatory semantic features and integration of the resulting quantities into a radiomics framework may improve the overall interpretability of radiomics from a clinical-physiological perspective. For example, in a 2021 paper by Choi et al. [ 135 ], an interpretable spiculation feature based on spiculation quantification was used for radiomics modeling. The model achieved an AUC of 0.82 on the Lung Image Database Consortium dataset and an AUC of 0.76 on a LUNGx dataset. In 2022, Choi et al. [ 136 ] released a large-scale dataset, the Clinically-Interpretable Radiomics Dataset, which focuses on features on the surface of pulmonary nodules such as spiculation or lobulation and sharp or curved spikes, as these features can be clinically explained. For example, the formation of spiculation can be explained by the proliferation of fibrous connective tissue caused by the infiltration of or stimulation of surrounding tumor cells.

Statistical modeling

After feature engineering, a suitable model is developed based on the selected features. Researchers can choose a single machine learning algorithm to generate a model and then evaluate its performance, or use several algorithms to generate various models and then compare their performance to identify the best model. It is currently believed that no algorithm is the best in all scenarios, so researchers must choose the appropriate algorithm for a given scenario, which may be an SVM [ 137 ], a logistic regression [ 138 ], a KNN [ 139 ], a decision tree (DT) [ 140 ], a RF [ 141 ], or an extreme gradient boosting (XGBoost) algorithm [ 142 ]. In recent years, due to advancements in deep learning technology, researchers have increasingly used deep learning methods to construct models.

We summarize the strengths and limitations of some algorithms and classify them into machine learning methods, deep learning methods and statistical methods. In the second part of this paper, we extract various algorithms from the references. Table 9 focuses on some common algorithms in these references. Statistical methods are often used to evaluate the data differences between training sets and validation sets, and the differences between survey groups. In addition, Kaplan–Meier analysis and log-rank test are generally combined to compare the OS of different risk groups, so these two methods generally appear simultaneously in prognosis research. The application of machine learning methods is diverse, including feature selection, model construction, model performance validation and so on. About feature selection, it has been described in detail in the subsection: Feature selection of feature engineering. Some algorithms are both common in feature selection and model construction, such as LASSO and SVM. Traditional machine learning methods usually divide the problem to be solved into several sub-problems and then solve them one by one. Deep learning methods solve problems in an end-to-end way, which means the feature engineering step is not required. However, correspondingly, deep learning methods are less interpretive and have higher requirements for machines and equipment.

Researchers typically use the following metrics or methods to evaluate model performance: AUC, Acc, F-1 score, sensitivity, specificity, precision, positive and negative predictive values, calibration curve analysis, decision curve analysis, the Hosmer–Lemeshow test, or the Akaike information criterion. These reveal certain characteristics of a model, such as its discriminability, generalizability, robustness, clinical utility, and goodness of fit. Nevertheless, many aspects remain to be improved in radiomics models. Here, we review four of these aspects.

Model generalizability

The generalizability of a model is its ability to predict unknown data. The smaller the deviation of a model-predicted result from the real result, the higher the generalizability of the model. Thus, a model with high generalizability is better than a model with low generalizability. Accordingly, generalizability must be considered when radiomics models are applied in clinical decision making.

There are many reasons why a radiomics model may exhibit low generalizability. The root cause may be a difference between the distribution of the training data and that of the unknown data. For example, as there is currently no standard workflow for radiomics, changes in any aspect of a workflow, such as data sources, scanners, acquisition protocols, or image segmentation methods, will cause some level of data discrepancy that will affect the performance of a model. Castillo et al. [ 143 ] found that radiomics models had an excellent ability to distinguish between low- and high-grade prostate tumors when single-center data or single-scanner data were used, but this ability was significantly reduced when multicenter data and/or multi-scanner data were used. Therefore, standardizing these processes can reduce variability and improve generalizability. However, as it will take some years for this to be achieved, another approach must currently be used: examining the number and diversity of input samples. Therefore, researchers have attempted to develop models using multicenter data and to circumvent the ethical and legal problems raised by multicenter data sharing by using distributed learning, which aims to train radiomics models without requiring the data to “leave” the hospital where it is housed. For example, Bogowicz et al. [ 144 ] constructed centralized and distributed HNC survival-prediction radiomics models and found that these models exhibit no significant AUC differences in terms of feature selection and classification. This confirmed that distributed learning does not affect model performance and indicates that it is a valid new approach for adopting multicenter data to generate models with good generalizability.

When a large amount of data cannot be obtained, generalizability can be improved by small sample learning and manifold learning. Small sample learning incorporates data augmentation and transfer learning [ 145 ]: data augmentation increases sample size and diversity [ 146 ], while transfer learning allows knowledge to be learned from related domains to increase model performance in the target domain. A study published in 2022 [ 147 ] described a feature extraction method based on transfer learning that increases the number of samples, suggested that the deep radiomics features extracted by this method might be more accurate than manually extracted radiomics features, and then confirmed this by comparing the performance of models for predicting the regression of early HNC. Transfer learning involves domain adaptation: the adjustment of a model to enable it to adapt to different domains (i.e., different datasets), so the model exhibits good generalizability (i.e., can be applied to unfamiliar sample sets) [ 148 ]. For example, Wang et al. [ 149 ] performed cross-phase adversarial domain adaptation using a gastric CT arterial phase as the source domain and the pre-contrast phase as the target domain, and thereby obtained a radiomics model that performed well in predicting the target domain. In addition, manifold learning can increase the number of samples by reducing the dimensionality of high-dimensional image data. Accordingly, Zhang et al. [ 150 ] developed a classifier called multi-kernel regression with graph embedding, which applies manifold learning to prevent a radiomics model overfitting when predicting distant metastases of nasopharyngeal carcinoma. The classifier embeds a class compactness graph and connects a pair of samples of the same class in low-dimensional label space via undirected weighted edges to understand their relationship. Then, it maximizes the closeness between samples of the same class to prevent overfitting and enhance generalizability.

Researchers have also improved models’ generalizability by employing appropriate feature-selection approaches. Shayesteh et al. [ 151 ] used the mutual information feature-selection method to enhance the generalizability of radiomics models; specifically, they analyzed non-linear relationships and linear relationships to find radiomics features with the highest discriminability, which they used to enhance model performance. As a target dataset may be nearly identical or completely different from an original dataset, the repeatability and reproducibility of features are the factors influencing a model’s generalizability. Therefore, enhancing the reproducibility and reproducibility of radiomics features can also enhance a model’s generalizability. As such, the harmonization schemes are crucial, such as ComBat. For example, a study published in 2022 [ 152 ] demonstrated that specific post-processing algorithms can be applied to coordinate PET image noise and thereby increase the agreement of radiomics features. Another study [ 153 ] developed convolutional neural networks (CNNs) to coordinate CT images with different reconstruction kernels, which helps to improve a feature’s reproducibility and thus a model’s generalizability. Moreover, data augmentation is an image coordination method. Ibrahim et al. [ 154 ] adopted the radiomics feature harmonization scheme of ComBat harmonization and found that it improved feature agreement when the acquisition and reconstruction parameters were significantly different. However, they also noted that direct application of the ComBat method was not invariably successful, such that pre-adjustment of the parameters of the phantom dataset was necessary. Image coordination has also been achieved by identifying robust features. Robinson et al. [ 155 ] found that a model’s generalizability decreased monotonously with a reduction in feature reproducibility, meaning that the classification generalizability of the model was improved by selecting radiomics features with high robustness.

Several studies have been performed to select features with high repeatability and reproducibility for use in feature engineering. Lu et al. [ 99 ] developed a new phantom-based framework to screen radiomics features for repeatability and reproducibility and identify robust features by evaluating the effects of biological and noise signals. A study [ 156 ] published in 2021 used a new method (which differs from embedded methods) for selecting robust features for predicting the mutation status of isocitrate dehydrogenase 1/2 (IDH1/2) in glioma. In this method, feature selection involves the identification of variables that are part of causal structures (based on causal reasoning), are insensitive to environmental changes, are highly robust, and have a constant relationship with the IDH1/2 mutation status. Radiomics models were constructed based on these variables, and their generalizability and performance could be improved even in a set of non-observational environments.

Imbalanced data classification

In an imbalanced data set, there are large differences between the sizes of different classes of data. The generalizability of a radiomics model based on such a data set will be affected by these differences, which may decrease its predictive power. That is, in a case of a class imbalance, a model first learns that there is more of one class of data than another in the training set, so the model’s predictions are biased toward the majority class [ 157 ]. For example, if there are two classes of data, class 1 and class 2, and they are present in a ratio of 1:100, then the predictive Acc of a model based on class 2 data may be much higher than class 1 data. However, if we mainly focus on the 1 class, the model will not meet our requirements.

Resampling techniques are essential for dealing with class imbalances and have been used in many radiomics studies. These techniques increase the balance of an imbalanced dataset at the data level and involve either over-sampling or under-sampling. The former is performed by copying the minority class data to add more data [ 158 ], while the latter is performed by discarding a large amount of the majority class data. Regarding under-sampling techniques, a radiomics study mentioned above [ 149 ] developed the so-called sensibly based under-sampling imbalanced integrated framework, which divides the majority class data into different blocks by clustering data on the basis of sample sensitivity. This under-sampling technique comprises two steps. In the first step, a coordinated method is used to maintain the same sensitivity level in each block. In the second step, a self-paced factor is applied to reduce the sample probability of the block with a large sample size and is combined with multi-kernel regression with graph embedding to train a good radiomics model. Regarding over-sampling techniques, a study that categorized pathologic complete responses (pCRs) for cancer [ 159 ] included 222 patients (61 pCR and 161 non-pCR patients) in its training set. The two types of patient samples were significantly different, so the researchers applied the synthetic minority over-sampling technique (SMOTE) algorithm to an MRI-based radiomics feature set to balance the minority and majority differences between the samples. Although SMOTE is an over-sampling technology, it synthesizes similar new samples from original samples rather than copying original samples. In addition, Zhang et al. [ 160 ] developed a novel over-sampling network—DeepSMOTE, and pioneered the integration of convolutional image features with radiomics features to effectively enhance the classification ability of an unbalanced dataset. The target data for this study were rare rim-positive lesions present in quantitative susceptibility mapping. DeepSMOTE increases these data by adding the two nearest neighbors of each rim-positive lesion, and then linearly combines the features of each lesion and its nearest-neighbor to generate comprehensive data.

Researchers have also adopted algorithm-based techniques to address data set class-imbalance problems and thus alleviate the degradation of radiomics model performance. Ensemble learning, for example, combines several weak algorithms to obtain a more comprehensive and strong algorithm that performs bagging, boosting, and stacking, and studies have confirmed that ensemble learning-generated algorithms can be well applied to imbalanced data [ 161 , 162 ]. In 2022, Tang et al. [ 163 ] reported the advantages of bagging ensemble learning for prognostic prediction of HNSCC and showed that a bagging ensemble radiomics model generated more accurate predictions (an average Acc of up to 88.3%) than DT, RF, XGBoost, SVM, and linear models. Another study [ 164 ] compared the predictive performance of four DT and ensemble radiomics models based on boosting with that of an SVM model and found that the performance of the former models was superior to that of the latter model. Cost-sensitivity learning is also an effective approach that can yield excellent results by increasing the cost of algorithmic misclassification [ 165 ]. In 2021, Sun et al. [ 166 ] combined cost-sensitivity learning with ensemble learning, as this is effective for application to imbalanced datasets. Moreover, a tiny class can be detected and then treated as an anomaly. For example, Welch et al. [ 167 ] used isolation forest algorithms to detect abnormal data and solve a data set class-imbalance problem when applying two pipelines (i.e., machine learning and deep learning pipelines) and synthesizing patient-specific features, clinical features, radiomics features, and quantitative radiation therapy features to predict the local failure of HNC treatment.

Furthermore, researchers have combined data-level approaches with algorithm-based techniques. Jiao et al. [ 168 ] extracted the radiomics features of US images and then constructed a cost-sensitive SVM model, a SMOTE and adaptive boosting (AdaBoost) combination model, and a random under-sampling and AdaBoost combination model (RUSBoost), for the original unbalanced data set. They also built a SVM model and an AdaBoost model and applied them to the original unbalanced data set and an adaptive synthetic-nominal (ADASYN) algorithm-balanced data set, respectively. All of the models obtained good results, with the RUSBoost model exhibiting the best performance.

Multi-modality fusion

The use of multi-modalities is an unsolved challenge in many fields, including in radiomics, where researchers must decide what modes to incorporate into studies and when to do so [ 169 ]. Many radiomics studies have adopted multi-modality fusion approaches and have concluded that a radiomics model based on multi-modal information is superior to a radiomics model based on single-modality information. Generally, there are three time nodes for multi-modality fusion, namely early fusion, intermediate fusion and late fusion [ 170 ], and the advantages of each are described in the “Limitations and suggestions” section.

Early fusion is also known as data-level fusion and is the fusion of multiple modalities’ information before a feature input classifier is implemented [ 171 ]. Li et al. [ 172 ] combined the MR image features of different sequences to construct radiomics signatures and then combined these with clinic-radiological risk factors to develop a multi-factor model based on a training set. They found that the multi-factor model had the highest performance. Another study [ 173 ] modeled radiomics features with the tumor–node–metastasis stage primary tumor volume, clinical and biological features, respectively, and found that the performance of the multi-factor model was better than that of the single-factor model. Sheikh et al. [ 174 ] compared a CT model, an MRI model, and a CT–MRI hybrid model, and found that the hybrid model achieved the highest AUC in predicting acute radiation-induced xerostomia in HNC, as their respective AUCs were 0.57, 0.66 and 0.70 in the external validation cohort.

Intermediate fusion is also known as inter-layer fusion as it is the fusion of modalities between the input and output layer during modeling [ 175 ]. A study published in 2021 [ 120 ] explored the ability of multi-omics models to predict the eligibility of patients with nasopharyngeal carcinoma for adaptive radiation therapy by using multi-kernel learning algorithms to achieve intermediate fusion of four types of omics features: radiomics, dosiomics, contouromics, and morphology features. Specifically, four single-omics models (a radiomics model, a dosiomics model, a contouromics model, and a morphology model) and four multi-omics models (a radiomics–dosiomics model, a radiomics–contouromics model, a radiomics–morphology model, and a radiomics–morphology–dosiomics–contouromics model) were constructed and studied, and decision graphs were used to select which model was the best, i.e., which combination of the above-mentioned omics features formed the highest-performing model. The results revealed the superiority of the radiomics features: the AUC of the radiomics model (0.94) was the highest of the four single-omics models, and radiomics features comprised the largest proportion of features in the multi-omics models. These results also indicated that the performance of multi-omics models was generally better than that of single-omics models.

Late fusion is also known as decision-level fusion as it involves building a model by performing a certain processing fusion of different modalities to enhance the model’s performance. Chen et al. [ 176 ] adopted a late fusion approach to devise a mixed predictive model. That is, they established a novel many-objective radiomics model and a 3D-CNN model, and then applied an evidence-reasoning method to fuse the outputs of these two models to obtain a higher predictive Acc than that obtained from either model alone. They also fused two kinds of image data: PET data and CT data. Their results revealed that the mixed model input with PET and CT data exhibited a predictive Acc superior to that of a single model input with only CT or PET data.

Other multi-modality fusions can be employed, such as radiomics feature and gene signature fusion or radiomics features fusion based on PET and MRI data. Researchers have yet to fully explore all forms of multi-modality fusions, so this remains a fruitful and exciting avenue of enquiry.

Model interpretability

The “black box” characteristics of the machine learning approaches used to construct radiomics models, combined with the high sensitivity of radiomics features to image-specific variations [ 177 ], mean that radiomics models can be complex. Moreover, most radiomics studies have been single-center retrospective studies and thus had inherent defects such as small sample sizes and ineffectively verified model generalizability. Furthermore, the decision-making processes of deep learning models are opaque and may be unknowable. All of the above-mentioned aspects mean that radiomics models may have low interpretability, i.e., the results of models cannot be explained and so are not trusted by doctors. In these cases, the clinical implementation and development of models are severely hindered.

There is a growing body of research focusing on solving interpretability problems. Thus far, most of this research has explored either the local interpretation of specific predictions or global interpretation of working principles [ 178 ]. In machine learning, linear models and DT algorithms are inherently globally interpretable. In contrast, the local interpretable model-agnostic explanations (LIME) technique makes use of prediction samples and perturbed samples generated by random perturbation to fit a simple interpretable model, which provides local explanations for a black-box model. Zafar et al. [ 179 ] constructed a deterministic LIME framework that uses hierarchical clustering and a KNN algorithm instead of random perturbations to enhance the stability of explanations. In addition, shapley additive explanations (SHAP) and LIME have often been compared, and SHAP has been a common method of interpretation. For example, Giraud et al. [ 180 ] used SHAP when developing a radiomics model for oropharyngeal cancer, determined SHAP values to quantify the contribution of each feature to predicting local recurrence, and used an interpretable method to identify the most important risk factors.

Partial dependence plots (PDPs) are applied for the global interpretation of working principles as they express the relationship between prediction targets and variables (features), which renders a black box model visible and thereby effectively increases interpretability [ 181 ]. Accumulated local effects (ALE) plots are a superior alternative to PDP, and an ALE plot was used by Tan et al. [ 182 ] to reveal the major effects of each radiomics feature they examined. Their ALE plot demonstrated that higher “GreyLevelNonUniformity” values reflected intratumoral heterogeneity, while lower “Strength” values and more image-defined risk factors were associated with a higher probability of MYCN oncogene amplification. Permutation importance is also a key technique, whereby features are constantly adjusted during the testing of a model and the importance of all of the features to the predictive ability of the model is evaluated by observing changes in the performance of the model [ 183 ]. Enke et al. [ 184 ] applied permutation importance to screen the most relevant predictive radiomics features from 100 ranked radiomics features. They concluded that the features on the Laplacian of a Gaussian-filtered image were the most important when all of the features were included, while the shape features were the most important when only the radiomics features of the original image were included.

One study [ 185 ] simplified selected radiomics features to enhance the interpretability of a model. It used only 5 of 42 related features and the whole omental tumor volume to build a model; this involved discarding most of the uninterpretable features, thus making the model easier to interpret than a model that included all of the features. Moreover, this simplified model maintained a high predictive power (AUC: 0.68 ± 0.03) on the external test set. Another study [ 186 ] used a gene masking technique to improve physicians’ acceptance of a model. Specifically, the gene expression profiles of patients were obtained, and the expression of related genes was retained while that of unrelated genes was masked. Then, the masked expression data was input into the radiomics model, and the predictive performance of the model was calculated. This process was repeated for the entire cohort, with a higher performance indicating a stronger predictive correlation between the gene set and the radiomics features. The above mentioned study in 2020 [ 132 ] also increased interpretability to produce a radiomics feature-activation map, which revealed that peritumoral regions were more relevant than GTVs for distinguishing histological subtypes of NSCLC in CT imaging.

The interpretability of black-box models can be improved via many other methods, but these have yet to be applied in radiomics. We hope that future research will increase the diversity of methods available to solve the interpretability problems of radiomics models.

Summary and analysis

In this review, we review research on the application of radiomics for clinical diagnosis, prognosis, and determination of treatment responses, as well as research examining the two most important steps in radiomics analysis: feature engineering and statistical modeling.

Feature engineering consists of feature extraction and feature selection. Feature extraction can be performed to extract 2D or 3D image features, and there is no consensus on whether either type of features is superior. Each type has its own advantages and can be selected according to the imaging method used or the nature of the research being performed. In addition, researchers have attempted to identify new radiomics signatures. Feature selection is conducted to reduce the redundancy between features and improve the correlation between features and clinical goals. Feature selection methods comprise filter methods, embedding methods, and wrapper methods. The latter methods exhibit overfitting problems and thus have rarely been used in radiomics research, and none of the remaining methods are suitable for all situations. The reproducibility of radiomics features is a key concern in radiomics research. Reproducibility is sensitive to external factors, and all of the steps preceding radiomics modeling can affect feature reproducibility. The reproducibility of features can be investigated by following the radiomics workflow: image acquisition, image preprocessing, delineation of tumor ROIs, and feature selection. Features are extremely sensitive to the changes in acquisition details; even two images of the same tissue site may be different due to differences between their acquisition details. Nevertheless, in various acquisition modes and with certain reconstruction parameters, some features exhibit good reproducibility and therefore are good candidates for tumor segmentation. Inter- and intra-scanner feature reproducibility is affected by the slice thickness during image acquisition: thick slices are best for measuring tumor volumes and changes in these volumes, as thin slices contain comparatively more noise, which obscures texture features. Various methods are used for image preprocessing; a common method is resampling of voxel size, which can solve the problem of different voxel sizes to enhance the reproducibility of radiomics features. In addition, the ComBat harmonization method can eliminate most scanner effects and improve feature reproducibility and thus is suitable for multicenter research. Although manual segmentation is the preferred method for delineating tumor ROIs, it may be replaced gradually by fully automatic and semi-automatic segmentation due to the continual improvements in the Acc of the latter two methods. The reproducibility of manual segmentation can be improved by its being performed by several experts, followed by inter-expert evaluation. Contouring protocols or related guidelines are recommended for reducing inter-expert variability in target volume delineation. Some studies have devised new feature selection methods to obtain radiomics features with high reproducibility. Radiomics features can be clearly defined at a mathematical level, which adds some level of interpretability to radiomics results, but radiomics features’ lack of interpretability at the biological level limits the application and development of radiomics in medicine.

The generalizability of models has received additional attention, as it determines whether models can be applied in real-world multicenter scenarios. The root cause of generalization problems is the differences between the distribution of training samples and that of testing samples, which originate at every step of the radiomics workflow. Therefore, a standardized radiomics research plan must be determined in future research. At this stage, this problem can be solved by using techniques such as transfer learning and manifold learning. In addition, incorporating robust features into a model can effectively improve its generalizability because, compared with features with low robustness, features with high robustness are better able to resist environmental changes. Imbalanced datasets are another factor that adversely affects a model’s generalizability and predictive ability. This problem has often been solved by employing data-level resampling techniques, such as over-sampling and under-sampling. In addition, an algorithm-level imbalance processing strategy has exhibited some advantages, such as its ability to construct a more sensitive loss function and a more reasonable integration strategy than would have otherwise been available.

Furthermore, multi-modal fusion and model interpretability have also been the focus of some attention in radiomics modeling. Research has shown that multi-modal fusion methods encompass one of three time nodes: early fusion, intermediate fusion, and late fusion. Regardless of the fusion method used, studies have revealed that compared with single-modal fusion, multi-modal fusion can obtain more mode information and thereby afford better prediction results. For example, a multivariate model that combines radiomics features with clinical factors was shown to have better predictive performance than a radiomics features-only model or a clinical factors-only model. Analogously, the performance of fusion models based on radiomics, doseomics, and contouromics was shown to be better than the performance of models based on only a single type of omics. In addition, combining traditional radiomics features with other features (such as deep learning features or genetic features) and combining different imaging methods (such as combining CT imaging and PET imaging) have revealed new pathways in radiomics research.

Ensuring the interpretability of a radiomics model is the last step prior to the model’s clinical application. Some studies have suggested that a model can be explained from a global or local perspective. LIME is used to build simple models to explain the local parts of a complex model; PDP and global interpretation techniques, such as ALE, are used to visualize complex models; and interpretation methods, such as SHAP and gene masking techniques, are used to quantify the relationship between features and diseases (or genes) and thereby increase the interpretability of a model.

Limitations and suggestions

Although the studies reviewed above have demonstrated that radiomics achieves good results when applied to tumors, many limitations hinder the broad application of radiomics models in real-world clinical settings. Some studies have attempted to alleviate these limitations, but current technology is not sufficiently advanced to completely eliminate them. Given these challenges, we now offer some insights and suggestions.

A large number of radiomics features can be derived from images in radiomics study. It is important to control the number of radiomics features because too many features can lead to overfitting. To the best of our knowledge, so far, there is no absolute rule that fits all scenarios to determine the number of radiomics features required for modeling. In radiomics-based study, the optimal number of features is often determined by cross-validation on the internal validation data. This cross-validation strategy is applicable to any scenario because it is training data-based. However, studies do show that there is a relationship between the number of features and the number of training samples through a large number of experiments. For example, Hua et al. [ 187 ] selected 7 classifiers and carried out extensive experiments to find the relationship between the number of features and the training sample size. They found that the behavior of the optimal-feature-size relative to the sample size depends strongly on the classifier and the feature-label distribution. An immediate corollary is that one should be wary of rules-of-thumb generalized from specific cases. In addition, the performance of a designed classifier can be greatly influenced by the number of features and therefore one should attempt to use a number close to the optimal number. This means that it can be useful to refer to a database of optimal-feature-size curves to choose a feature size, even if this means making a necessarily very coarse approximation of the distribution model from the data—even perhaps just a visual assessment of the data.

Many studies have explored the reproducibility of features in different situations, but it appears that no studies have systematically summarized features that are robust to various influencing factors. Most studies have focused on image segmentation when exploring the reproducibility of features. However, the reproducibility of manual segmentation is usually not high due to large inter-expert deviations in defined tumor boundaries. The features obtained by some automatic or semi-automatic segmentation algorithms are more robust than manual features, but manual segmentation has been the most common form of segmentation conducted in recent clinical research. This may be because researchers have paid more attention to Acc in small-scale studies, whereas radiomics results have been applied in clinical practice. The reproducibility of segmentation may be more important than its Acc, as was noted by Kumar et al. [ 188 ]. Moreover, even if manual segmentation is replaced by automatic segmentation in the future, this does not mean that the same type of automatic segmentation will be applicable in all situations. Thus, it may be necessary to select an appropriate automatic segmentation algorithm for a given research application and set of parameters. Similarly, no consensus has been reached on the optimal ICC threshold, but it is often the preferred indicator for reproducibility analysis. Therefore, it may be apposite for future research to identify an appropriate ICC threshold value.

Test–retest analysis is often conducted for dimensionality reduction and selection of robust radiomics features with minimal changes, but test–retest analysis results may not be generalizable, and it is recommended that conditions specific to treatment sites and time, scanners, and imaging protocols are used [ 189 ]. However, test–retest may be impractical as its utility may not be understood or accepted by patients in a clinic. The phantom study is not described too much in this article for feature stability and repeatability studies. Phantom study can improve the reproducibility and robustness of radiomics features, while the phantom can also be used to accurately measure multicenter differences between different scanners or environments. The reproducibility of radiomics features based on CT is affected by material-dependence [ 190 ], which means that it is very important to select appropriate phantom materials according to the characteristics of different tumor sites and ensure that the range of features value between them is similar. Although phantom studies have shown promising trends in relevant radiomics analyses, it remains doubtful whether the phantom study results can be transferred to clinical studies. Mackin et al. [ 191 ] partially answered this question; they found that the variability in the values of radiomics features in phantom CT images was comparable to the variability in the values of radiomics features in NSCLC tumor CT images.

Poor model generalizability is one of the main reasons why models have failed to be widely adopted in clinical practice. Methods that have been used to improve generalizability are discussed in this study, and studies have revealed that no single method is applicable to all scenarios. It is not enough to extract robust features to improve model generalizability. Oliveira et al. [ 192 ] stated that it is necessary to use standardized multicenter datasets for radiomics research, and they found that a model constructed based on a standardized multicenter dataset had better predictive performance (AUC: 0.67–0.74) than a predictive model constructed based on robust features in validation cohorts (AUC = 0.53). Nevertheless, another study [ 193 ] came to the opposite conclusion: the features of multicenter MR images exhibited significantly more variations than those of single-center MR images. Most of the variations were related to the differences in hardware and acquisition, which can influence apparent diffusion coefficient diagrams. The authors provided ways to correct the data variations, such as by discarding poorly reproducible features, performing normalization, and using statistical models that specifically take center effects into account. The authors also mentioned that the reproducibility of high-order radiomics features was poor, which suggests that researchers should carefully select high-order features when modeling. Furthermore, technologies such as harmonization or few-shot learning are not optimal solutions; even though researchers try their best to improve generalizability, the resulting models cannot be perfectly applied to all target datasets. To sum up, to improve model performance on external validation datasets, the solutions can be organized into three categories: data-level, structure-level and algorithm-level. At the data-level, multicenter data standardizing can be used to reduce the distribution difference between training data and external validation data. In structure-level, centralized or distributed learning structure can make multicenter studies close to single-center studies. At algorithm-level, transfer learning can leverage knowledge from target-related domains to train a model on a target task. Manifold learning aims to use regularization terms to minimize the distribution differences between training data and external validation data. In real application scenarios, to the best of our knowledge, so far, there is no absolute rule for determining which strategy can be used to improve model performance on external validation datasets. With the development of federated learning, perhaps structure-level will become more favored.

Radiomics features are highly sensitive to various parameters, and every step in the radiomics workflow must be considered when using additional methods to attempt to improve the stability features, which increases the cost of radiomics analysis. Therefore, data sharing and pipeline standardization are the only way to reduce the variation in radiomics data and improve the generalizability of models. However, it remains difficult to standardize radiomics workflows and implement these at all institutions. Therefore, a model’s generalizability should be improved as much as possible, and then the model should be externally validated to improve its reliability so that it can be applied to clinical decision making.

Imbalanced datasets are common in learning models based on tumor images (as hospitals or institutions store too few samples of target patients), and this is a major limitation of radiomics. This can be addressed by two types of sampling methods, as summarized in a study [ 194 ] that contrasted these methods’ effects on radiomics model performance with that of classifier tuning methods and feature selection algorithms. This showed that an optimal choice of classifiers and feature selection algorithms could significantly improve model performance but did not show that sampling methods had a great impact on model performance. Thus, algorithm optimization methods are better than sampling methods in this context. However, Xie et al. [ 195 ] argued that model parameter tuning and feature selection did not significantly improve the Acc of a radiomics model. Instead, they found that over-sampling techniques such as ADASYN and SMOTE can raise the geometric means and F-measures of minority class data of HNC patients, thereby improving the predictive performance of imbalanced datasets. Therefore, we propose that to maximize the performance of a model, the best modeling method for the given research dataset must be comprehensively determined.

As multi-modality fusion has always been an obstacle in radiomics, and multi-modality-based models often generate better results than single-modality-based models, multi-modality fusion has strong potential for improving radiomics models and is a current research hotspot. If imbalanced classes are present in a data set, a classifier is biased towards the majority class when learning and thus generates incorrect predictions, which affects the robustness of the model. This is a difficult problem that must be solved, and two types of solutions have been devised: data-based improvements and algorithm-based improvements. Multi-modality can be a combination of radiomics features with other types of features or radiomics with other omics, or a combination of multiple imaging methods, and refinement can result in a combination of different MR imaging sequences [ 196 ]. It is well understood that multi-modality methods can compensate for the shortcomings of single-modality methods. Early fusion can use different types of data for complementary purposes. For example, the preferred imaging method for tumors is CT, but this has insufficient soft-tissue contrast; however, the addition of MRI or PET information to a CT data set can overcome this problem to some extent [ 197 ]. Late fusion improves final results by fusing models’ results. It has the obvious advantage that model errors do not affect each other and thus errors do not accumulate. Intermediate fusion can perform multiple fusions between layers, which further explores the complementarity between modalities that is explored via early fusion. In summary, this explains why multi-factor models are generally superior to single-factor models. Thus, when conducting fusion research, researchers should perform numerous experiments with different settings and predictive models to enable careful selection of fusion modality and timing, as this will enhance the quality of the resulting model. In reviewing the literature, we found that even though the number of studies using multi-modality fusion has increased in recent years, it remains a rare approach overall and requires further attention from researchers.

Radiomics, due to its high complexity and use of black-box machine learning, has the problem of low interpretability in both its features and models. This problem has caused physicians and specialists to distrust radiomics and has been the biggest obstacle to the widespread adoption of radiomics models in clinical settings. Researchers have thus developed methods to increase the interpretability of radiomics features and models. At the feature level, exploring the association between features and tumor heterogeneity can increase interpretability. At the model level, diverse technologies based on local and global interpretation can be applied to improve the interpretability of a model, although such technologies have their own shortcomings. For example, a feature map formed by deep convolutional neural networks can allow the most uninterpretable deep learning fields to realize the visual interpretation of classification tasks based on these networks [ 198 ]. However, we found that not all of the techniques for interpreting black-box models have been thoroughly studied in the field of radiomics, so this remains a task for future research. In addition, as most of the learning of radiomics models is retrospective and not invariably persuasive, it remains insufficient to use only interpretative models, as no model is applicable to all clinical decision-making scenarios. Researchers must continue to perform prospective studies to verify the clinical utility of the methods described in this review.

Conclusions

In this review, we introduce the radiomics-based studies in cancers from the perspectives of clinical applications and AI-driven modeling. In the first perspective, we provide a particular focus on three distinct applications: tumor grading, tumor staging, and the classification of benign vs. malignant tumors. In the second perspective, we devote more text to discussing feature engineering and statistical modeling in AI-driven radiomics modeling, including feature reproducibility, feature interpretability, model generalizability, model interpretability, imbalanced data classification, and multi-modality fusion. Our comprehensive review reveals that, in the context of AI, radiomics-based studies indeed play an important role in the diagnosis and prognostic prediction of cancers. However, related studies on feature reproducibility, feature interpretability, model generalization and model interpretability still present challenges, which are obstacles to the further promotion of radiomics models to clinical real-world applications.

Availability of data and materials

Not applicable.

Abbreviations

All-cause mortality

Adaptive boosting

Adaptive synthetic-nominal

Artificial intelligence

American Joint Committee on Cancer

Accumulated local effects

Artificial neural network

Analysis of variance

Area under the receiver operating characteristic curve

Concordance correlation coefficient

Convolutional neural network

Coefficients of variation

Computed tomography

Cross validation

Disease-free survival

Deep learning

Distant metastasis

Decision tree

Extra-nodal extension

Fibrosing interstitial lung disease

Grey level co-occurrence matrix

Gross tumor volumes

Hepatocellular carcinoma

Head and neck cancer

Head and neck squamous cell carcinoma

Human papillomavirus

Induction chemotherapy

Intraclass correlation coefficients

K-nearest neighbors

Kernel principal component analysis

Least absolute shrinkage and selection operator

Local interpretable model-agnostic explanations

Lymph node metastasis

Leave one out cross validation

Logistic regression

Mutual information

Machine learning

Magnetic resonance imaging

Minimal redundancy maximal relevance

Metabolic tumor volume

Neoadjuvant chemotherapy

Neighborhood component analysis

Nasopharyngeal carcinoma

Non-small cell lung cancer

Oropharyngeal cancer

Oropharyngeal squamous cell carcinoma

Overall survival

Principal component analysis

Pathologic complete response

Partial dependence plots

Positron emission tomography

Progression-free survival

Planning target volumes

Quantitative US

Rim average

Random forest

Recursive feature elimination

Region of interest

Random under-sampling and AdaBoost combination model

Shapley additive explanations

Statistical method

Synthetic minority over-sampling technique

Solitary pulmonary lesions

Support-vector machine

Ultrasonography

Variance-threshold

Extreme gradient boosting

Two-dimensional

Three-dimensional

Ferlay J, Colombet M, Soerjomataram I, Mathers C, Parkin DM, Piñeros M, et al. Estimating the global cancer incidence and mortality in 2018: GLOBOCAN sources and methods. Int J Cancer. 2019;144(8):1941–53.

Article CAS PubMed Google Scholar

Yin L, Li Z, Xue G, Lu L. Nasopharyngeal carcinoma misdiagnosed as pituitary tumor with multiple cranial neuropathies. J Cancer Res Ther. 2021;17(7):1748–50.

Article PubMed Google Scholar

Hou Z, Li S, Ren W, Liu J, Yan J, Wan S. Radiomic analysis in T2W and SPAIR T2W MRI: predict treatment response to chemoradiotherapy in esophageal squamous cell carcinoma. J Thorac Dis. 2018;10(4):2256–67.

Article PubMed PubMed Central Google Scholar

Zhang L, Li L, Feng G, Fan T, Jiang H, Wang Z. Advances in CT techniques in vascular calcification. Front Cardiovasc Med. 2021;8:716822.

Chabanova E, Larsen L, Løgager VB, Møller JM, Thomsen HS. Use of magnetic resonance imaging. Ugeskr Laeger. 2014;176(1):50–4.

PubMed Google Scholar

Zhu A, Marcus DM, Shu HK, Shim H. Application of metabolic PET imaging in radiation oncology. Radiat Res. 2012;177(4):436–48.

Article CAS PubMed PubMed Central Google Scholar

Liu Z, Wang S, Dong D, Wei J, Fang C, Zhou X, et al. The applications of radiomics in precision diagnosis and treatment of oncology: opportunities and challenges. Theranostics. 2019;9(5):1303–22.

Lambin P, Leijenaar RTH, Deist TM, Peerlings J, De Jong EEC, Van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14(12):749–62.

Scapicchio C, Gabelloni M, Barucci A, Cioni D, Saba L, Neri E. A deep look into radiomics. Radiol Med. 2021;126(10):1296–311.

Lim H, Kim DH, Jung HY, Gong EJ, Na HK, Ahn JY, et al. Clinical significance of early detection of esophageal cancer in patients with head and neck cancer. Gut Liver. 2015;9(2):159–65.

Mydlarz WK, Hennessey PT, Califano JA. Advances and perspectives in the molecular diagnosis of head and neck cancer. Expert Opin Med Diagn. 2010;4(1):53–65.

Yip SS, Aerts HJ. Applications and limitations of radiomics. Phys Med Biol. 2016;61(13):R150–66.

Wu W, Ye J, Wang Q, Luo J, Xu S. CT-based radiomics signature for the preoperative discrimination between head and neck squamous cell carcinoma grades. Front Oncol. 2019;9:821.

Mukherjee P, Cintra M, Huang C, Zhou M, Zhu S, Colevas AD, et al. CT-based radiomic signatures for predicting histopathologic features in head and neck squamous cell carcinoma. Radiol Imaging Cancer. 2020;2(3):e190039.

Parmar C, Leijenaar RT, Grossmann P, Rios Velazquez E, Bussink J, Rietveld D, et al. Radiomic feature clusters and prognostic signatures specific for lung and head & neck cancer. Sci Rep. 2015;5:11044.

Wang F, Zhang B, Wu X, Liu L, Fang J, Chen Q, et al. Radiomic nomogram improves preoperative T category accuracy in locally advanced laryngeal carcinoma. Front Oncol. 2019;9:1064.

Ren J, Tian J, Yuan Y, Dong D, Li X, Shi Y, et al. Magnetic resonance imaging based radiomics signature for the preoperative discrimination of stage I-II and III-IV head and neck squamous cell carcinoma. Eur J Radiol. 2018;106:1–6.

Wu L, Wang C, Tan X, Cheng Z, Zhao K, Yan L, et al. Radiomics approach for preoperative identification of stages I-II and III-IV of esophageal cancer. Chin J Cancer Res. 2018;30(4):396–405.

Liang C, Huang Y, He L, Chen X, Ma Z, Dong D, et al. The development and validation of a CT-based radiomics signature for the preoperative discrimination of stage I–II and stage III–IV colorectal cancer. Oncotarget. 2016;7(21):31401–12.

Gao RZ, Wen R, Wen DY, Huang J, Qin H, Li X, et al. Radiomics analysis based on ultrasound images to distinguish the tumor stage and pathological grade of bladder cancer. J Ultrasound Med. 2021;40(12):2685–97.

Ho TY, Chao CH, Chin SC, Ng SH, Kang CJ, Tsang NM. Classifying neck lymph nodes of head and neck squamous cell carcinoma in MRI images with radiomic features. J Digit Imaging. 2020;33(3):613–8.

Yin J, Qiu JJ, Qian W, Ji L, Yang D, Jiang JW, et al. A radiomics signature to identify malignant and benign liver tumors on plain CT images. J Xray Sci Technol. 2020;28(4):683–94.

Piludu F, Marzi S, Ravanelli M, Pellini R, Covello R, Terrenato I, et al. MRI-based radiomics to differentiate between benign and malignant parotid tumors with external validation. Front Oncol. 2021;11:656918.

Zhang S, Yu M, Chen D, Li P, Tang B, Li J. Role of MRI-based radiomics in locally advanced rectal cancer (review). Oncol Rep. 2022;47(2):34.

Nowakowski A, Lahijanian Z, Panet-Raymond V, Siegel PM, Petrecca K, Maleki F, et al. Radiomics as an emerging tool in the management of brain metastases. Neurooncol Adv. 2022;4(1):vdac141.

PubMed PubMed Central Google Scholar

Shalata AT, Shehata M, Van Bogaert E, Ali KM, Alksas A, Mahmoud A, et al. Predicting recurrence of non-muscle-invasive bladder cancer: current techniques and future trends. Cancers (Basel). 2022;14(20):5019.

Li M, Zhu YZ, Zhang YC, Yue YF, Yu HP, Song B. Radiomics of rectal cancer for predicting distant metastasis and overall survival. World J Gastroenterol. 2020;26(33):5008–21.

Fh T, Cyw C, Eyw C. Radiomics AI prediction for head and neck squamous cell carcinoma (HNSCC) prognosis and recurrence with target volume approach. BJR Open. 2021;3(1):20200073.

Folkert MR, Setton J, Apte AP, Grkovski M, Young RJ, Schöder H, et al. Predictive modeling of outcomes following definitive chemoradiotherapy for oropharyngeal cancer based on FDG-PET image characteristics. Phys Med Biol. 2017;62(13):5327–43.

Haider SP, Zeevi T, Baumeister P, Reichel C, Sharaf K, Forghani R, et al. Potential added value of PET/CT radiomics for survival prognostication beyond AJCC 8th edition staging in oropharyngeal squamous cell carcinoma. Cancers (Basel). 2020;12(7):1778.

Woolen S, Virkud A, Hadjiiski L, Cha K, Chan HP, Swiecicki P, et al. Prediction of disease free survival in laryngeal and hypopharyngeal cancers using CT perfusion and radiomic features: a pilot study. Tomography. 2021;7(1):10–9.

Bologna M, Corino V, Calareso G, Tenconi C, Alfieri S, Iacovelli NA, et al. Baseline MRI-radiomics can predict overall survival in non-endemic EBV-related nasopharyngeal carcinoma patients. Cancers (Basel). 2020;12(10):2958.

Liu K, Qiu Q, Qin Y, Chen T, Zhang D, Huang L, et al. Radiomics nomogram based on multiple-sequence magnetic resonance imaging predicts long-term survival in patients diagnosed with nasopharyngeal carcinoma. Front Oncol. 2022;12:852348.

Wang F, Tan R, Feng K, Hu J, Zhuang Z, Wang C, et al. Magnetic resonance imaging-based radiomics features associated with depth of invasion predicted lymph node metastasis and prognosis in tongue cancer. J Magn Reson Imaging. 2022;56(1):196–209.

Li H, Chen XL, Liu H, Lu T, Li ZL. MRI-based multiregional radiomics for predicting lymph nodes status and prognosis in patients with resectable rectal cancer. Front Oncol. 2022;12:1087882.

M. D. Anderson Cancer Center Head and Neck Quantitative Imaging Working Group. Investigation of radiomic signatures for local recurrence using primary tumor texture analysis in oropharyngeal head and neck cancer patients. Sci Rep. 2018;8(1):1524.

Article Google Scholar

Kim M, Lee JH, Joo L, Jeong B, Kim S, Ham S, et al. Development and validation of a model using radiomics features from an apparent diffusion coefficient map to diagnose local tumor recurrence in patients treated for head and neck squamous cell carcinoma. Korean J Radiol. 2022;23(11):1078–88.

Dasgupta A, Bhardwaj D, Dicenzo D, Fatima K, Osapoetra LO, Quiaoit K, et al. Radiomics in predicting recurrence for patients with locally advanced breast cancer using quantitative ultrasound. Oncotarget. 2021;12(25):2437–48.

Zhang L, Dong D, Li H, Tian J, Ouyang F, Mo X, et al. Development and validation of a magnetic resonance imaging-based model for the prediction of distant metastasis before initial treatment of nasopharyngeal carcinoma: a retrospective cohort study. EBioMedicine. 2019;40:327–35.

Sharma D, Koshy G, Grover S, Sharma B. Sentinel lymph node biopsy: a new approach in the management of head and neck cancers. Sultan Qaboos Univ Med J. 2017;17(1):e3–10.

Li F, Pan D, He Y, Wu Y, Peng J, Li J, et al. Using ultrasound features and radiomics analysis to predict lymph node metastasis in patients with thyroid cancer. BMC Surg. 2020;20(1):315.

Lin M, Tang X, Cao L, Liao Y, Zhang Y, Zhou J. Using ultrasound radiomics analysis to diagnose cervical lymph node metastasis in patients with nasopharyngeal carcinoma. Eur Radiol. 2023;33(2):774–83.

Cysouw MCF, Jansen BHE, Van De Brug T, Oprea-Lager DE, Pfaehler E, De Vries BM, et al. Machine learning-based analysis of [ 18 F] DCFPyL PET radiomics for risk stratification in primary prostate cancer. Eur J Nucl Med Mol Imaging. 2021;48(2):340–9.

Tran WT, Suraweera H, Quiaoit K, Dicenzo D, Fatima K, Jang D, et al. Quantitative ultrasound delta-radiomics during radiotherapy for monitoring treatment responses in head and neck malignancies. Future Sci OA. 2020;6(9):FSO624.

Kaźmierska J, Kaźmierski MR, Bajon T, Winiecki T, Bandurska-Luque A, Ryczkowski A, et al. Prediction of incomplete response of primary tumour based on clinical and radiomics features in inoperable head and neck cancers after definitive treatment. J Pers Med. 2022;12(7):1092.

Zhang MH, Cao D, Ginat DT. Radiomic model predicts lymph node response to induction chemotherapy in locally advanced head and neck cancer. Diagnostics (Basel). 2021;11(4):588.

Xi Y, Ge X, Ji H, Wang L, Duan S, Chen H, et al. Prediction of response to induction chemotherapy plus concurrent chemoradiotherapy for nasopharyngeal carcinoma based on MRI radiomics and delta radiomics: a two-center retrospective study. Front Oncol. 2022;12:824509.

Chen BY, Xie H, Li Y, Jiang XH, Xiong L, Tang XF, et al. MRI-based radiomics features to predict treatment response to neoadjuvant chemotherapy in locally advanced rectal cancer: a single center, prospective study. Front Oncol. 2022;12:801743.

Guo L, Du S, Gao S, Zhao R, Huang G, Jin F, et al. Delta-radiomics based on dynamic contrast-enhanced MRI predicts pathologic complete response in breast cancer patients treated with neoadjuvant chemotherapy. Cancers (Basel). 2022;14(14):3515.

Xiang Y, Li S, Wang H, Song M, Hu K, Wang F, et al. MRI-based radiomics to predict neoadjuvant chemoradiotherapy outcomes in locally advanced rectal cancer: a multicenter study. Clin Transl Radiat Oncol. 2023;38:175–82.

King AD, Griffith JF, Abrigo JM, Leung SF, Yau FK, Tse GM, et al. Osteoradionecrosis of the upper cervical spine: MR imaging following radiotherapy for nasopharyngeal carcinoma. Eur J Radiol. 2010;73(3):629–35.

Zhong X, Li L, Jiang H, Yin J, Lu B, Han W, et al. Cervical spine osteoradionecrosis or bone metastasis after radiotherapy for nasopharyngeal carcinoma? The MRI-based radiomics for characterization. BMC Med Imaging. 2020;20(1):104.

Van Dijk LV, Thor M, Steenbakkers R, Apte A, Zhai TT, Borra R, et al. Parotid gland fat related magnetic resonance image biomarkers improve prediction of late radiation-induced xerostomia. Radiother Oncol. 2018;128(3):459–66.

Dercle L, Zhao B, Gönen M, Moskowitz CS, Connors DE, Yang H, et al. An imaging signature to predict outcome in metastatic colorectal cancer using routine computed tomography scans. Eur J Cancer. 2022;161:138–47.

Jiang Y, Wang H, Wu J, Chen C, Yuan Q, Huang W, et al. Noninvasive imaging evaluation of tumor immune microenvironment to predict outcomes in gastric cancer. Ann Oncol. 2020;31(6):760–8.

Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. 2016;278(2):563–77.

Haralick RM, Shanmugam K, Dinstein I. Textural features for image classification. TSMC. 1973;SMC-3(6):610–21.

Google Scholar

Galloway MM. Texture analysis using gray level run lengths. Comput Graph Image Proc. 1975;4(2):172–9.

Sun C, Wee WG. Neighboring gray level dependence matrix for texture classification. Comput Vis Graph Image Proc. 1983;23(3):341–52.

Shen C, Liu Z, Guan M, Song J, Lian Y, Wang S, et al. 2D and 3D CT radiomics features prognostic performance comparison in non-small cell lung cancer. Transl Oncol. 2017;10(6):886–94.

Wan Q, Zhou J, Xia X, Hu J, Wang P, Peng Y, et al. Diagnostic performance of 2D and 3D T2WI-based radiomics features with machine learning algorithms to distinguish solid solitary pulmonary lesion. Front Oncol. 2021;11:683587.

Xu L, Yang P, Yen EA, Wan Y, Jiang Y, Cao Z, et al. A multi-organ cancer study of the classification performance using 2D and 3D image features in radiomics analysis. Phys Med Biol. 2019;64(21):215009.

Houseni M, Mahmoud MA, Saad S, Elhussiny F, Shihab M. Advanced intra-tumoural structural characterisation of hepatocellular carcinoma utilising FDG-PET/CT: a comparative study of radiomics and metabolic features in 3D and 2D. Pol J Radiol. 2021;86:e64–73.

Zhu Y, Yao W, Xu BC, Lei YY, Guo QK, Liu LZ, et al. Predicting response to immunotherapy plus chemotherapy in patients with esophageal squamous cell carcinoma using non-invasive radiomic biomarkers. BMC Cancer. 2021;21(1):1167.

Beichel RR, Ulrich EJ, Smith BJ, Bauer C, Brown B, Casavant T, et al. FDG PET based prediction of response in head and neck cancer treatment: assessment of new quantitative imaging features. PLoS ONE. 2019;14(4):e0215465.

Buizza G, Toma-Dasu I, Lazzeroni M, Paganelli C, Riboldi M, Chang Y, et al. Early tumor response prediction for lung cancer patients using novel longitudinal pattern features from sequential PET/CT image scans. Phys Med. 2018;54:21–9.

Van Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, et al. Computational radiomics system to decode the radiographic phenotype. Cancer Res. 2017;77(21):e104–7.

Fedorov A, Beichel R, Kalpathy-Cramer J, Finet J, Fillion-Robin JC, Pujol S, et al. 3D Slicer as an image computing platform for the quantitative imaging network. Magn Reson Imaging. 2012;30(9):1323–41.

Zhang L, Fried DV, Fave XJ, Hunter LA, Yang J, Court LE. IBEX: an open infrastructure software platform to facilitate collaborative work in radiomics. Med Phys. 2015;42(3):1341–53.

Zwanenburg A, Vallières M, Abdalah MA, Aerts H, Andrearczyk V, Apte A, et al. The image biomarker standardization initiative: standardized quantitative radiomics for high-throughput image-based phenotyping. Radiology. 2020;295(2):328–38.

Avanzo M, Wei L, Stancanello J, Vallières M, Rao A, Morin O, et al. Machine and deep learning methods for radiomics. Med Phys. 2020;47(5):e185–202.

Xie C, Yang P, Zhang X, Xu L, Wang X, Li X, et al. Sub-region based radiomics analysis for survival prediction in oesophageal tumours treated by definitive concurrent chemoradiotherapy. EBioMedicine. 2019;44:289–97.

Peng H, Long F, Ding C. Feature selection based on mutual information: criteria of max-dependency, max-relevance, and min-redundancy. IEEE Trans Pattern Anal Mach Intell. 2005;27(8):1226–38.

Hu C, Zheng D, Cao X, Pang P, Fang Y, Lu T, et al. Application value of magnetic resonance radiomics and clinical nomograms in evaluating the sensitivity of neoadjuvant chemotherapy for nasopharyngeal carcinoma. Front Oncol. 2021;11:740776.

Kira K, Rendell LA. A practical approach to feature selection. In: Sleeman D, Edwards P, editors. Proceedings of the ninth international workshop on machine learning. Aberdeen: Morgan Kaufmann Publishers Inc; 1992. p. 249–56.

Student. The probable error of a mean. Biometrika. 1908;6:1–25.

Jin X, Xu A, Bie R, Guo P. Machine learning techniques and Chi-square feature selection for cancer classification using SAGE gene expression profiles. In: Li J, Yang Q, Tan A-H, editors. Data mining for biomedical applications. Berlin: Springer; 2006. p. 106–15.

Chapter Google Scholar

Parmar C, Grossmann P, Bussink J, Lambin P, Aerts H. Machine learning methods for quantitative radiomic biomarkers. Sci Rep. 2015;5:13087.

Koul N, Manvi SS. Ensemble feature selection from cancer gene expression data using mutual information and recursive feature elimination. In: 2020 third international conference on advances in electronics, computers and communications (ICAECC). Bengaluru: IEEE; 2020. p. 1–6.

Yu Q, Wang A, Gu J, Li Q, Ning Y, Peng J, et al. Multiphasic CT-based radiomics analysis for the differentiation of benign and malignant parotid tumors. Front Oncol. 2022;12:913898.

Tibshirani R. Regression shrinkage and selection via the LASSO: a retrospective. J R Stat Soc B. 2011;73(3):273–82.

Duan W, Xiong B, Tian T, Zou X, He Z, Zhang L. Radiomics in nasopharyngeal carcinoma. Clin Med Insights Oncol. 2022;16:11795549221079186.

Li Y, Yu M, Wang G, Yang L, Ma C, Wang M, et al. Contrast-enhanced CT-based radiomics analysis in predicting lymphovascular invasion in esophageal squamous cell carcinoma. Front Oncol. 2021;11:644165.

Song F, Guo Z, Mei D. Feature selection using principal component analysis. In: 2010 International conference on system science, engineering design and manufacturing informatization. Yichang: IEEE; 2010. p. 27–30.

Van Der Maaten L, Hinton G. Viualizing data using t-SNE. J Mach Learn Res. 2008;9:2579–605.

Teng X, Zhang J, Ma Z, Zhang Y, Lam S, Li W, et al. Improving radiomic model reliability using robust features from perturbations for head-and-neck carcinoma. Front Oncol. 2022;12:974467.

Teng X, Zhang J, Zwanenburg A, Sun J, Huang Y, Lam S, et al. Building reliable radiomic models using image perturbation. Sci Rep. 2022;12(1):10035.

Jensen LJ, Kim D, Elgeti T, Steffen IG, Hamm B, Nagel SN. Stability of radiomic features across different region of interest sizes-a CT and MR phantom study. Tomography. 2021;7(2):238–52.

Flouris K, Jimenez-Del-Toro O, Aberle C, Bach M, Schaer R, Obmann MM, et al. Assessing radiomics feature stability with simulated CT acquisitions. Sci Rep. 2022;12(1):4732.

Balagurunathan Y, Kumar V, Gu Y, Kim J, Wang H, Liu Y, et al. Test-retest reproducibility analysis of lung CT image features. J Digit Imaging. 2014;27(6):805–23.

Midya A, Chakraborty J, Gönen M, Do RKG, Simpson AL. Influence of CT acquisition and reconstruction parameters on radiomic feature reproducibility. J Med Imaging (Bellingham). 2018;5(1):011020.

Tu SJ, Chen WY, Wu CT. Uncertainty measurement of radiomics features against inherent quantum noise in computed tomography imaging. Eur Radiol. 2021;31(10):7865–75.

Michallek F, Genske U, Niehues SM, Hamm B, Jahnke P. Deep learning reconstruction improves radiomics feature stability and discriminative power in abdominal CT imaging: a phantom study. Eur Radiol. 2022;32(7):4587–95.

Yan J, Chu-Shern JL, Loi HY, Khor LK, Sinha AK, Quek ST, et al. Impact of image reconstruction settings on texture features in 18F-FDG PET. J Nucl Med. 2015;56(11):1667–73.

Galavis PE, Hollensen C, Jallow N, Paliwal B, Jeraj R. Variability of textural features in FDG PET images due to different acquisition modes and reconstruction parameters. Acta Oncol. 2010;49(7):1012–6.

Prayer F, Hofmanninger J, Weber M, Kifjak D, Willenpart A, Pan J, et al. Variability of computed tomography radiomics features of fibrosing interstitial lung disease: a test-retest study. Methods. 2021;188:98–104.

Zhao B, Tan Y, Tsai WY, Qi J, Xie C, Lu L, et al. Reproducibility of radiomics for deciphering tumor phenotype with imaging. Sci Rep. 2016;6:23428.

Haarburger C, Müller-Franzes G, Weninger L, Kuhl C, Truhn D, Merhof D. Radiomics feature reproducibility under inter-rater variability in segmentations of CT images. Sci Rep. 2020;10(1):12688.

Lu L, Sun SH, Afran A, Yang H, Lu ZF, So J, et al. Identifying robust radiomics features for lung cancer by using in-vivo and phantom lung lesions. Tomography. 2021;7(1):55–64.

Meyer M, Ronald J, Vernuccio F, Nelson RC, Ramirez-Giraldo JC, Solomon J, et al. Reproducibility of CT radiomic features within the same patient: influence of radiation dose and CT reconstruction settings. Radiology. 2019;293(3):583–91.

Fiset S, Welch ML, Weiss J, Pintilie M, Conway JL, Milosevic M, et al. Repeatability and reproducibility of MRI-based radiomic features in cervical cancer. Radiother Oncol. 2019;135:107–14.

Rai R, Holloway LC, Brink C, Field M, Christiansen RL, Sun Y, et al. Multicenter evaluation of MRI-based radiomic features: a phantom study. Med Phys. 2020;47(7):3054–63.

Shafiq-Ul-Hassan M, Zhang GG, Latifi K, Ullah G, Hunt DC, Balagurunathan Y, et al. Intrinsic dependencies of CT radiomic features on voxel size and number of gray levels. Med Phys. 2017;44(3):1050–62.

Linsalata S, Borgheresi R, Marfisi D, Barca P, Sainato A, Paiar F, et al. Radiomics of patients with locally advanced rectal cancer: effect of preprocessing on features estimation from computed tomography imaging. Biomed Res Int. 2022;2022:2003286.

Shafiq-Ul-Hassan M, Latifi K, Zhang G, Ullah G, Gillies R, Moros E. Voxel size and gray level normalization of CT radiomic features in lung cancer. Sci Rep. 2018;8(1):10545.

Jensen LJ, Kim D, Elgeti T, Steffen IG, Schaafs LA, Hamm B, et al. Enhancing the stability of CT radiomics across different volume of interest sizes using parametric feature maps: a phantom study. Eur Radiol Exp. 2022;6(1):43.

Bologna M, Corino V, Mainardi L. Technical Note: virtual phantom analyses for preprocessing evaluation and detection of a robust feature set for MRI-radiomics of the brain. Med Phys. 2019;46(11):5116–23.

Li Y, Ammari S, Balleyguier C, Lassau N, Chouzenoux E. Impact of preprocessing and harmonization methods on the removal of scanner effects in brain MRI radiomic features. Cancers (Basel). 2021;13(12):3000.

Mali SA, Ibrahim A, Woodruff HC, Andrearczyk V, Müller H, Primakov S, et al. Making radiomics more reproducible across scanner and imaging protocol variations: a review of harmonization methods. J Pers Med. 2021;11(9):842.

Orlhac F, Lecler A, Savatovski J, Goya-Outi J, Nioche C, Charbonneau F, et al. How can we combat multicenter variability in MR radiomics? Validation of a correction procedure. Eur Radiol. 2021;31(4):2272–80.

Orlhac F, Eertink JJ, Cottereau AS, Zijlstra JM, Thieblemont C, Meignan M, et al. A guide to ComBat harmonization of imaging biomarkers in multicenter studies. J Nucl Med. 2022;63(2):172–9.

Orlhac F, Boughdad S, Philippe C, Stalla-Bourdillon H, Nioche C, Champion L, et al. A postreconstruction harmonization method for multicenter radiomic studies in PET. J Nucl Med. 2018;59(8):1321–8.

Wang L, Tan J, Ge Y, Tao X, Cui Z, Fei Z, et al. Assessment of liver metastases radiomic feature reproducibility with deep-learning-based semi-automatic segmentation software. Acta Radiol. 2021;62(3):291–301.

Parmar C, Rios Velazquez E, Leijenaar R, Jermoumi M, Carvalho S, Mak RH, et al. Robust radiomics feature quantification using semiautomatic volumetric segmentation. PLoS ONE. 2014;9(7):e102107.

Pavic M, Bogowicz M, Würms X, Glatz S, Finazzi T, Riesterer O, et al. Influence of inter-observer delineation variability on radiomics stability in different tumor sites. Acta Oncol. 2018;57(8):1070–4.

Gitto S, Bologna M, Corino VDA, Emili I, Albano D, Messina C, et al. Diffusion-weighted MRI radiomics of spine bone tumors: feature stability and machine learning-based classification performance. Radiol Med. 2022;127(5):518–25.

Denzler S, Vuong D, Bogowicz M, Pavic M, Frauenfelder T, Thierstein S, et al. Impact of CT convolution kernel on robustness of radiomic features for different lung diseases and tissue types. Br J Radiol. 2021;94(1120):20200947.

Bologna M, Corino VDA, Montin E, Messina A, Calareso G, Greco FG, et al. Assessment of stability and discrimination capacity of radiomic features on apparent diffusion coefficient images. J Digit Imaging. 2018;31(6):879–94.

Yan KK, Wang X, Lam WWT, Vardhanabhuti V, Lee AWM, Pang HH. Radiomics analysis using stability selection supervised component analysis for right-censored survival data. Comput Biol Med. 2020;124:103959.

Lam SK, Zhang Y, Zhang J, Li B, Sun JC, Liu CY, et al. Multi-organ omics-based prediction for adaptive radiation therapy eligibility in nasopharyngeal carcinoma patients undergoing concurrent chemoradiotherapy. Front Oncol. 2021;11:792024.

Leijenaar RT, Carvalho S, Velazquez ER, Van Elmpt WJ, Parmar C, Hoekstra OS, et al. Stability of FDG-PET radiomics features: an integrated analysis of test–retest and inter-observer variability. Acta Oncol. 2013;52(7):1391–7.

Hu Y, Xie C, Yang H, Ho JWK, Wen J, Han L, et al. Computed tomography-based deep-learning prediction of neoadjuvant chemoradiotherapy treatment response in esophageal squamous cell carcinoma. Radiother Oncol. 2021;154:6–13.

Wang J, Zeng J, Li H, Yu X. A deep learning radiomics analysis for survival prediction in esophageal cancer. J Healthc Eng. 2022;2022:4034404.

Paul R, Schabath M, Balagurunathan Y, Liu Y, Li Q, Gillies R, et al. Explaining deep features using radiologist-defined semantic features and traditional quantitative features. Tomography. 2019;5(1):192–200.

Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006.

Rifi AL, Dufait I, Aisati CE, Ridder MD, Barbé K. Unraveling the biological meaning of radiomic features. In: 2022 IEEE international symposium on medical measurements and applications (MeMeA). Messina: IEEE; 2022. p. 1–6.

Just N. Improving tumour heterogeneity MRI assessment with histograms. Br J Cancer. 2014;111(12):2205–13.

Eloyan A, Yue MS, Khachatryan D. Tumor heterogeneity estimation for radiomics in cancer. Stat Med. 2020;39(30):4704–23.

Wang X, Xie T, Luo J, Zhou Z, Yu X, Guo X. Radiomics predicts the prognosis of patients with locally advanced breast cancer by reflecting the heterogeneity of tumor cells and the tumor microenvironment. Breast Cancer Res. 2022;24(1):20.

Ferreira Junior JR, Koenigkam-Santos M, Machado CVB, Faleiros MC, Correia NSC, Cipriano FEG, et al. Radiomic analysis of lung cancer for the assessment of patient prognosis and intratumor heterogeneity. Radiol Bras. 2021;54(2):87–93.

Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, Van Stiphout RG, Granton P, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441–6.

Vuong D, Tanadini-Lang S, Wu Z, Marks R, Unkelbach J, Hillinger S, et al. Radiomics feature activation maps as a new tool for signature interpretability. Front Oncol. 2020;10:578895.

Kuthuru S, Deaderick W, Bai H, Su C, Vu T, Monga V, et al. A visually interpretable, dictionary-based approach to imaging-genomic modeling, with low-grade glioma as a case study. Cancer Inform. 2018;17:1176935118802796.

Luo Y, Mcshan D, Ray D, Matuszak M, Jolly S, Lawrence T, et al. Development of a fully cross-validated Bayesian network approach for local control prediction in lung cancer. IEEE Trans Radiat Plasma Med Sci. 2019;3(2):232–41.

Choi W, Nadeem S, Alam SR, Deasy JO, Tannenbaum A, Lu W. Reproducible and interpretable spiculation quantification for lung cancer screening. Comput Methods Programs Biomed. 2021;200:105839.

Choi W, Dahiya N, Nadeem S. CIRDataset: a large-scale dataset for clinically-interpretable lung nodule radiomics and malignancy prediction. Med Image Comput Comput Assist Interv. 2022;2022:13–22.

Chen J, Jiao L. Classification mechanism of support vector machines. WCC 2000—ICSP 2000. In: 2000 5th international conference on signal processing proceedings. 16th world computer congress 2000; 2000 21–25 Aug; Beijing: IEEE; 2000. p. 1556–9.

Cramer JS. The origins of logistic regression. SSRN Electron J. 2002. https://doi.org/10.2139/ssrn.360300 .

Saini I, Singh D, Khosla A. QRS detection using k-nearest neighbor algorithm (KNN) and evaluation on standard ECG databases. J Adv Res. 2013;4(4):331–44.

Quinlan JR. Induction of decision trees. Mach Learn. 1986;1(1):81–106.

Breiman L. Random forests. Mach Learn. 2001;45(1):5–32.

Chen T, Guestrin C. XGBoost: a scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. San Francisco: Association for Computing Machinery; 2016. p. 785–94.

Castillo TJ, Starmans MPA, Arif M, Niessen WJ, Klein S, Bangma CH, et al. A multi-center, multi-vendor study to evaluate the generalizability of a radiomics model for classifying prostate cancer: high grade vs. low grade. Diagnostics (Basel). 2021;11(2):369.

Bogowicz M, Jochems A, Deist TM, Tanadini-Lang S, Huang SH, Chan B, et al. Privacy-preserving distributed learning of radiomics to predict overall survival and HPV status in head and neck cancer. Sci Rep. 2020;10(1):4542.

Wang Y, Yao Q, Kwok J, Ni L. Generalizing from a few examples: a survey on few-shot learning. ACM Comput Surv. 2020;53:1–34.

Nalepa J, Marcinkiewicz M, Kawulok M. Data augmentation for brain-tumor segmentation: a review. Front Comput Neurosci. 2019;13:83.

Tanaka S, Kadoya N, Sugai Y, Umeda M, Ishizawa M, Katsuta Y, et al. A deep learning-based radiomics approach to predict head and neck tumor regression for adaptive radiotherapy. Sci Rep. 2022;12(1):8899.

Guan H, Liu M. Domain adaptation for medical image analysis: a survey. IEEE Trans Biomed Eng. 2022;69(3):1173–85.

Wang S, Dong D, Li H, Feng C, Wang Y, Tian J. Cross-phase adversarial domain adaptation for deep disease-free survival prediction with gastric cancer CT images. Annu Int Conf IEEE Eng Med Biol Soc. 2021;2021:3501–4.

Zhang Y, Lam S, Yu T, Teng X, Zhang J, Lee FKH, et al. Integration of an imbalance framework with novel high-generalizable classifiers for radiomics-based distant metastases prediction of advanced nasopharyngeal carcinoma. Knowl-Based Syst. 2022;235:107649.

Shayesteh SP, Shiri I, Karami AH, Hashemian R, Kooranifar S, Ghaznavi H, et al. Predicting lung cancer patients’ survival time via logistic regression-based models in a quantitative radiomic framework. J Biomed Phys Eng. 2020;10(4):479–92.

Keller H, Shek T, Driscoll B, Xu Y, Nghiem B, Nehmeh S, et al. Noise-based image harmonization significantly increases repeatability and reproducibility of radiomics features in PET images: a phantom study. Tomography. 2022;8(2):1113–28.

Yoon JH, Sun SH, Xiao M, Yang H, Lu L, Li Y, et al. Convolutional neural network addresses the confounding impact of CT reconstruction kernels on radiomics studies. Tomography. 2021;7(4):877–92.

Ibrahim A, Refaee T, Leijenaar RTH, Primakov S, Hustinx R, Mottaghy FM, et al. The application of a workflow integrating the variable reproducibility and harmonizability of radiomic features on a phantom dataset. PLoS ONE. 2021;16(5):e0251147.

Robinson K, Li H, Lan L, Schacht D, Giger M. Radiomics robustness assessment and classification evaluation: a two-stage method demonstrated on multivendor FFDM. Med Phys. 2019;46(5):2145–56.

Santinha J, Matos C, Figueiredo M, Papanikolaou N. Improving performance and generalizability in radiogenomics: a pilot study for prediction of IDH1/2 mutation status in gliomas with multicentric data. J Med Imaging (Bellingham). 2021;8(3):031905.

Ramyachitra D, Manikandan P. Imbalanced dataset classification and solutions: a review. IJCBR. 2014;5(4):1–29.

Haibo H, Yang B, Garcia EA, Shutao L. ADASYN: adaptive synthetic sampling approach for imbalanced learning. In: 2008 IEEE international joint conference on neural networks (IEEE world congress on computational intelligence); 2008 1–8 Jun. Hong Kong: IEEE; 2008. p. 1322–8.

Sutton EJ, Onishi N, Fehr DA, Dashevsky BZ, Sadinski M, Pinker K, et al. A machine learning model that classifies breast cancer pathologic complete response on MRI post-neoadjuvant chemotherapy. Breast Cancer Res. 2020;22(1):57.

Zhang H, Nguyen TD, Zhang J, Marcille M, Spincemaille P, Wang Y, et al. QSMRim-Net: imbalance-aware learning for identification of chronic active multiple sclerosis lesions on quantitative susceptibility maps. Neuroimage Clin. 2022;34:102979.

Li K, Zhou G, Zhai J, Li F, Shao M. Improved PSO_AdaBoost ensemble algorithm for imbalanced data. Sensors (Basel). 2019;19(6):1476.

Khalilia M, Chakraborty S, Popescu M. Predicting disease risks from highly imbalanced data using random forest. BMC Med Inform Decis Mak. 2011;11:51.

Tang FH, Cheung EY, Wong HL, Yuen CM, Yu MH, Ho PC. Radiomics from various tumour volume sizes for prognosis prediction of head and neck squamous cell carcinoma: a voted ensemble machine learning approach. Life (Basel). 2022;12(9):1380.

Vamvakas A, Tsivaka D, Logothetis A, Vassiou K, Tsougos I. Breast cancer classification on multiparametric MRI—increased performance of boosting ensemble methods. Technol Cancer Res Treat. 2022;21:15330338221087828.

Pes B, Lai G. Cost-sensitive learning strategies for high-dimensional and imbalanced data: a comparative study. PeerJ Comput Sci. 2021;7:e832.

Sun Y, Li M, Li L, Shao H, Sun Y. Cost-sensitive classification for evolving data streams with concept drift and class imbalance. Comput Intell Neurosci. 2021;2021:8813806.

Welch ML, Mcintosh C, Mcniven A, Huang SH, Zhang BB, Wee L, et al. User-controlled pipelines for feature integration and head and neck radiation therapy outcome predictions. Phys Med. 2020;70:145–52.

Jiao J, Du Y, Li X, Guo Y, Ren Y, Wang Y. Prenatal prediction of neonatal respiratory morbidity: a radiomics method based on imbalanced few-shot fetal lung ultrasound images. BMC Med Imaging. 2022;22(1):2.

Dolmans TC, Poel M, Van’T Klooster JJR, Veldkamp BP. Perceived mental workload classification using intermediate fusion multimodal deep learning. Front Hum Neurosci. 2020;14:609096.

Ramachandram D, Taylor GW. Deep multimodal learning: a survey on recent advances and trends. IEEE Signal Proc Mag. 2017;34(6):96–108.

Zhang Y, Xia K, Jiang Y, Qian P, Cai W, Qiu C, et al. Multi-modality fusion & inductive knowledge transfer underlying non-sparse multi-kernel learning and distribution adaption. IEEE/ACM Trans Comput Biol Bioinform. 2022. https://doi.org/10.1109/tcbb.2022.3142748 .

Li W, Shen H, Han L, Liu J, Xiao B, Li X, et al. A multiparametric fusion radiomics signature based on contrast-enhanced MRI for predicting early recurrence of hepatocellular carcinoma. J Oncol. 2022;2022:3704987.

Keek SA, Wesseling FWR, Woodruff HC, Van Timmeren JE, Nauta IH, Hoffmann TK, et al. A prospectively validated prognostic model for patients with locally advanced squamous cell carcinoma of the head and neck based on radiomics of computed tomography images. Cancers (Basel). 2021;13(13):3271.

Sheikh K, Lee SH, Cheng Z, Lakshminarayanan P, Peng L, Han P, et al. Predicting acute radiation induced xerostomia in head and neck cancer using MR and CT radiomics of parotid and submandibular glands. Radiat Oncol. 2019;14(1):131.

Pan C, Zhou P, Tan J, Sun B, Guan R, Wang Z, et al . Liver tumor detection via a multi-scale intermediate multi-modal fusion network on MRI images. In: 2021 IEEE international conference on image processing (ICIP); 2021 19–22 Sep. Anchorage: IEEE; 2021. p. 299–303.

Chen L, Zhou Z, Sher D, Zhang Q, Shah J, Pham NL, et al. Combining many-objective radiomics and 3D convolutional neural network through evidential reasoning to predict lymph node metastasis in head and neck cancer. Phys Med Biol. 2019;64(7):075011.

Tiwari P, Verma R. The pursuit of generalizability to enable clinical translation of radiomics. Radiol Artif Intell. 2021;3(1):e200227.

Petch J, Di S, Nelson W. Opening the black box: the promise and limitations of explainable machine learning in cardiology. Can J Cardiol. 2022;38(2):204–13.

Zafar MR, Khan NMJA. DLIME: a deterministic local interpretable model-agnostic explanations approach for computer-aided diagnosis systems. ArXiv. 2019. https://doi.org/10.48550/arXiv.1906.10263 .

Giraud P, Giraud P, Nicolas E, Boisselier P, Alfonsi M, Rives M, et al. Interpretable machine learning model for locoregional relapse prediction in oropharyngeal cancers. Cancers (Basel). 2020;13(1):57.

Greenwell BM. pdp: an R package for constructing partial dependence plots. R J. 2017;9:421.

Tan E, Merchant K, Kn BP, Cs A, Zhao JJ, Saffari SE, et al. CT-based morphologic and radiomics features for the classification of MYCN gene amplification status in pediatric neuroblastoma. Childs Nerv Syst. 2022;38(8):1487–95.

Altmann A, Toloşi L, Sander O, Lengauer T. Permutation importance: a corrected feature importance measure. Bioinformatics. 2010;26(10):1340–7.

Enke JS, Moltz JH, D’anastasi M, Kunz WG, Schmidt C, Maurus S, et al. Radiomics features of the spleen as surrogates for CT-based lymphoma diagnosis and subtype differentiation. Cancers (Basel). 2022;14(3):713.

Rundo L, Beer L, Escudero Sanchez L, Crispin-Ortuzar M, Reinius M, Mccague C, et al. Clinically interpretable radiomics-based prediction of histopathologic response to neoadjuvant chemotherapy in high-grade serous ovarian carcinoma. Front Oncol. 2022;12:868265.

Smedley NF, Aberle DR, Hsu W. Using deep neural networks and interpretability methods to identify gene expression patterns that predict radiomic features and histology in non-small cell lung cancer. J Med Imaging (Bellingham). 2021;8(3):031906.

Hua J, Xiong Z, Lowey J, Suh E, Dougherty ER. Optimal number of features as a function of sample size for various classification rules. Bioinformatics. 2005;21(8):1509–15.

Kumar V, Gu Y, Basu S, Berglund A, Eschrich SA, Schabath MB, et al. Radiomics: the process and the challenges. Magn Reson Imaging. 2012;30(9):1234–48.

Van Timmeren JE, Leijenaar RTH, Van Elmpt W, Wang J, Zhang Z, Dekker A, et al. Test–retest data for radiomics feature stability analysis: generalizable or study-specific? Tomography. 2016;2(4):361–5.

Li Y, Reyhan M, Zhang Y, Wang X, Zhou J, Zhang Y, et al. The impact of phantom design and material-dependence on repeatability and reproducibility of CT-based radiomics features. Med Phys. 2022;49(3):1648–59.

Mackin D, Fave X, Zhang L, Fried D, Yang J, Taylor B, et al. Measuring computed tomography scanner variability of radiomics features. Investig Radiol. 2015;50(11):757–65.

Oliveira C, Amstutz F, Vuong D, Bogowicz M, Hüllner M, Foerster R, et al. Preselection of robust radiomic features does not improve outcome modelling in non-small cell lung cancer based on clinical routine FDG-PET imaging. EJNMMI Res. 2021;11(1):79.

Schurink NW, Van Kranen SR, Roberti S, Van Griethuysen JJM, Bogveradze N, Castagnoli F, et al. Sources of variation in multicenter rectal MRI data and their effect on radiomics feature reproducibility. Eur Radiol. 2022;32(3):1506–16.

Gabryś HS, Buettner F, Sterzing F, Hauswald H, Bangert M. Design and selection of machine learning methods using radiomics and dosiomics for normal tissue complication probability modeling of xerostomia. Front Oncol. 2018;8:35.

Xie C, Du R, Ho JW, Pang HH, Chiu KW, Lee EY, et al. Effect of machine learning re-sampling techniques for imbalanced datasets in 18 F-FDG PET-based radiomics model on prognostication performance in cohorts of head and neck cancer patients. Eur J Nucl Med Mol Imaging. 2020;47(12):2826–35.

Wang B, Yang J, Peng H, Ai J, An L, Yang B, et al. Brain tumor segmentation via multi-modalities interactive feature learning. Front Med (Lausanne). 2021;8:653925.

Daisne JF, Grégoire V. Multimodalities imaging for target volume definition in radiotherapy. Bull Cancer. 2006;93(12):1175–82.

Wang J, Yuan C, Han C, Wen Y, Lu H, Liu C, et al. IMAL-Net: interpretable multi-task attention learning network for invasive lung adenocarcinoma screening in CT images. Med Phys. 2021;48(12):7913–29.

Download references

Acknowledgements

This work was supported in part by the National Natural Science Foundation of China (82072019), the Shenzhen Basic Research Program (JCYJ20210324130209023), the Shenzhen-Hong Kong-Macau S&T Program (Category C) (SGDX20201103095002019), the Mainland-Hong Kong Joint Funding Scheme (MHKJFS) (MHP/005/20), the Project of Strategic Importance Fund (P0035421), the Projects of RISA (P0043001) from the Hong Kong Polytechnic University, the Natural Science Foundation of Jiangsu Province (BK20201441), the Provincial and Ministry Co-constructed Project of Henan Province Medical Science and Technology Research (SBGJ202103038, SBGJ202102056), the Henan Province Key R&D and Promotion Project (Science and Technology Research) (222102310015), the Natural Science Foundation of Henan Province (222300420575), and the Henan Province Science and Technology Research (222102310322).

Author information

Yuan-Peng Zhang, Xin-Yun Zhang, Yu-Ting Cheng, and Bing Li contributed equally to this work.

Authors and Affiliations

Department of Medical Informatics, Nantong University, Nantong, 226001, Jiangsu, China

Yuan-Peng Zhang, Xin-Yun Zhang & Yu-Ting Cheng

Department of Health Technology and Informatics, the Hong Kong Polytechnic University, Hong Kong, 999077, China

Yuan-Peng Zhang, Xin-Zhi Teng, Jiang Zhang, Saikit Lam, Ta Zhou, Zong-Rui Ma, Jia-Bao Sheng, Victor C. W. Tam, Shara W. Y. Lee & Jing Cai

The Hong Kong Polytechnic University Shenzhen Research Institute, Shenzhen, 518000, Guangdong, China

Yuan-Peng Zhang & Jing Cai

Department of Radiation Oncology, the Affiliated Cancer Hospital of Zhengzhou University and Henan Cancer Hospital, Zhengzhou, 450008, Henan, China

Bing Li & Hong Ge

You can also search for this author in PubMed Google Scholar

Contributions

YPZ, XYZ, YTC and BL did the literature review, drew figures, and drafted the manuscript; XZT, JZ, SL, TZ, ZRM, JBS, VCWT, SWYL and HG edited the manuscript; JC supervised the whole literature review. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Jing Cai .

Ethics declarations

Ethics approval and consent to participate, consent for publication, competing interests.

All the authors declare that they have no competing interest.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Zhang, YP., Zhang, XY., Cheng, YT. et al. Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling. Military Med Res 10 , 22 (2023). https://doi.org/10.1186/s40779-023-00458-8

Download citation

Received : 22 December 2022

Accepted : 04 May 2023

Published : 16 May 2023

DOI : https://doi.org/10.1186/s40779-023-00458-8

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Interpretability
Multi-modalities

Military Medical Research

ISSN: 2054-9369

Submission enquiries: Access here and click Contact Us
General enquiries: [email protected]

Global (EN)
Albania (en)
Algeria (fr)
Argentina (es)
Armenia (en)
Australia (en)
Austria (de)
Austria (en)
Azerbaijan (en)
Bahamas (en)
Bahrain (en)
Bangladesh (en)
Barbados (en)
Belgium (en)
Belgium (nl)
Bermuda (en)
Bosnia and Herzegovina (en)
Brasil (pt)
Brazil (en)
British Virgin Islands (en)
Bulgaria (en)
Cambodia (en)
Cameroon (fr)
Canada (en)
Canada (fr)
Cayman Islands (en)
Channel Islands (en)
Colombia (es)
Costa Rica (es)
Croatia (en)
Cyprus (en)
Czech Republic (cs)
Czech Republic (en)
DR Congo (fr)
Denmark (da)
Denmark (en)
Ecuador (es)
Estonia (en)
Estonia (et)
Finland (fi)
France (fr)
Georgia (en)
Germany (de)
Germany (en)
Gibraltar (en)
Greece (el)
Greece (en)
Hong Kong SAR (en)
Hungary (en)
Hungary (hu)
Iceland (is)
Indonesia (en)
Ireland (en)
Isle of Man (en)
Israel (en)
Ivory Coast (fr)
Jamaica (en)
Jordan (en)
Kazakhstan (en)
Kazakhstan (kk)
Kazakhstan (ru)
Kuwait (en)
Latvia (en)
Latvia (lv)
Lebanon (en)
Lithuania (en)
Lithuania (lt)
Luxembourg (en)
Macau SAR (en)
Malaysia (en)
Mauritius (en)
Mexico (es)
Moldova (en)
Monaco (en)
Monaco (fr)
Mongolia (en)
Montenegro (en)
Mozambique (en)
Myanmar (en)
Namibia (en)
Netherlands (en)
Netherlands (nl)
New Zealand (en)
Nigeria (en)
North Macedonia (en)
Norway (nb)
Pakistan (en)
Panama (es)
Philippines (en)
Poland (en)
Poland (pl)
Portugal (en)
Portugal (pt)
Romania (en)
Romania (ro)
Saudi Arabia (en)
Serbia (en)
Singapore (en)
Slovakia (en)
Slovakia (sk)
Slovenia (en)
South Africa (en)
Sri Lanka (en)
Sweden (sv)
Switzerland (de)
Switzerland (en)
Switzerland (fr)
Taiwan (en)
Taiwan (zh)
Thailand (en)
Trinidad and Tobago (en)
Tunisia (en)
Tunisia (fr)
Turkey (en)
Turkey (tr)
Ukraine (en)
Ukraine (ru)
Ukraine (uk)
United Arab Emirates (en)
United Kingdom (en)
United States (en)
Uruguay (es)
Uzbekistan (en)
Uzbekistan (ru)
Venezuela (es)
Vietnam (en)
Vietnam (vi)
Zambia (en)
Zimbabwe (en)
Financial Reporting View
Women's Leadership
Corporate Finance
Board Leadership
Executive Education

Fresh thinking and actionable insights that address critical issues your organization faces.

Insights by Industry
Insights by Topic

KPMG's multi-disciplinary approach and deep, practical industry knowledge help clients meet challenges and respond to opportunities.

Advisory Services
Audit Services
Tax Services

Services to meet your business goals

Technology Alliances

KPMG has market-leading alliances with many of the world's leading software and services vendors.

Helping clients meet their business challenges begins with an in-depth understanding of the industries in which they work. That’s why KPMG LLP established its industry-driven structure. In fact, KPMG LLP was the first of the Big Four firms to organize itself along the same industry lines as clients.

Our Industries

How We Work

We bring together passionate problem-solvers, innovative technologies, and full-service capabilities to create opportunity with every insight.

What sets us apart

Careers & Culture

What is culture? Culture is how we do things around here. It is the combination of a predominant mindset, actions (both big and small) that we all commit to every day, and the underlying processes, programs and systems supporting how work gets done.

Relevant Results

Sorry, there are no results matching your search., the role of managed services in the transformation agenda.

Navigating the ongoing journey with flexible operating models

In our era of compound volatility, it’s not enough to just transform. Many companies are doing so continually , with flexible operating models that keep evolving amid constant change.

One way to create those operating models is with managed services that are backed by strong advisory capabilities. Progressive organizations are using these services to facilitate ongoing transformation initiatives that are complex and interconnected.

To enable continuous transformation, KPMG recommends a focus on several key capabilities, and managed services can play an important role in each:

Clearly articulate the vision.

A transformation is a fundamental shift in the way an organization does business, and successful initiatives start with defining exactly how the change will create value—at the strategic, financial, and operational levels. These sources of value, ranging from reputation and revenue to process improvement and productivity, become the north star for an initiative.

In addition to defining what the initiative will achieve, consider what will be required. For example, in developing the business case for a cloud transformation, forward-thinking leaders explore not only the benefits that can come from software as a service (SaaS)—such as improvements in cost, speed, and performance—but also how to sustain them long term, beyond implementation. Many leaders establish a managed services model for ongoing optimization of the platform, helping them realize the business case.

In the transformation of a security operations center (SOC), as another example, information security leaders should consider how specifically the new SOC will enable the business—and how services will be delivered. The vision may include finding cyber threats faster, protecting critical assets, improving stakeholder trust, or securing new applications so they can be launched more quickly.

For a model that can continually deliver those benefits, with the flexibility to change course as needed, SOCs can collaborate with managed services providers in a shared responsibility model. These providers deploy advanced automation at scale while bringing skilled cyber practitioners to manage the technology.

Track value throughout the initiative.

In an environment of continuous transformation, it’s important not only to have a clear vision but also to decompose it into outcomes that can be measured along the way. This approach allows for course correction to mitigate risk.

For example, after the implementation of cloud-based recruiting software, measurable operational improvements could include:

Improved candidate experience
Reduced notification fatigue for recruiters
Improved efficiency of candidate applications
Faster administrative reporting on job requisitions
Improved workforce diversity due to analytics on ethnicity, gender, and age

These are the kinds of sustained improvements that can come from a managed services model for platform optimization.

Similarly, a SOC transformation with a managed services model could include measurable improvements such as:

Reduced cyber operational costs
Reduced adversary dwell time, demonstrating that the SOC can move at machine speed to quickly eradicate a threat
Improved customer and regulator trust

Faster speed to market for new applications, based on ongoing security monitoring and analytics in development processes

Manage ongoing change in multiple transformations.

According to recent KPMG research 1 , companies on average have three transformations running simultaneously, and a top reason for failure is inadequate change management. The increased pace and complexity of initiatives create a high risk of transformation burnout that can lead to high staff turnover.

To ease this transformation fatigue, companies can use managed services to build agile operating models that can flex up or down to meet fast-shifting priorities. Leading providers combine advanced technology, scale, and expertise to deliver key processes while evolving at the speed of business.

These providers also bring advisory services in change management, helping companies define the drivers for success—from policies and procedures to metrics and incentives—while aligning them across multiple transformations. For example, savvy providers know how to take a big-picture view, developing an enterprise-wide governance language to span overlapping initiatives.

Build a technology and data foundation to support ongoing transformation.

New technology is rarely a flip-the-switch solution that makes everything work. Technology can sometimes become a barrier that holds companies back – due to organizational disruption, poor adoption, lack of systems interoperability, or problems with the underlying data architecture.

Leading managed services providers, on the other hand, design operations that are tech-enabled but strategy-led. That means they bring not only sophisticated technology but also collaborative professionals with expertise in business functions, processes, industries, and data management. They help ensure that tech is an enabler, not a barrier.

For example, to continually enable a SaaS transformation, the best managed services providers assist with ongoing systems integration after implementation of the platform, while managing the impact that a change in one system has in another. Systems integration is also critical for shifting data to the cloud, where it can be made usable by generative AI.

Navigating the ongoing journey

In a world of constant flux, transformation is not a fixed destination, because by the time you get there, “there” has already moved on—with rapid market changes and new ways of doing business. That’s why many companies are trying to master the art of continuous transformation, and managed services are an important part of the playbook.

Learn more about KPMG Managed Services. For an archive of my past blogs, please visit Going Beyond: Managed Services.

1 The new transformation agenda , KPMG in the US

Explore more.

Managed Services: A new harmony of people and technology

Companies see managed services as part of their 'playbook' for sustained transformation

Opportunity is knocking. Are you answering?

How companies can collaborate with service providers to open the door to sustained digital transformation

Subscribe to Going Beyond: Managed Services

See our latest thinking on how managed services can help you drive transformation at the speed of business.

Meet our team

Explore other services tailored to your business

Thank you for contacting KPMG. We will respond to you as soon as possible.

Contact KPMG

By submitting, you agree that KPMG LLP may process any personal information you provide pursuant to KPMG LLP's Privacy Statement .

Job seekers

Visit our careers section or search our jobs database.

Use the RFP submission form to detail the services KPMG can help assist you with.

Office locations

International hotline

You can confidentially report concerns to the KPMG International hotline

Press contacts

Do you need to speak with our Press Office? Here's how to get in touch.

Home — Essay Samples — Life — My Heroes — My Mother Is My Role Model

Why My Role Model is My Mother

Categories: My Heroes Role Models Someone Who Inspires Me

About this sample

Words: 398 |

Published: Apr 29, 2022

Words: 398 | Page: 1 | 2 min read

My Role Model - an Essay About My Mother

Cite this Essay

Let us write you an essay from scratch

450+ experts on 30 subjects ready to help
Custom essay delivered in as few as 3 hours

Get high-quality help

Dr. Heisenberg

Verified writer

Expert in: Life

+ 120 experts online

By clicking “Check Writers’ Offers”, you agree to our terms of service and privacy policy . We’ll occasionally send you promo and account related email

No need to pay just yet!

Related Essays

1 pages / 612 words

2 pages / 984 words

1 pages / 593 words

1 pages / 569 words

Remember! This is just a sample.

You can get your custom paper by one of our expert writers.

121 writers online

Still can’t find what you need?

Browse our vast selection of original essay samples, each expertly formatted and styled

Related Essays on My Heroes

Along the way, as the world continues to revolve, our minds continue to grow around the world as well. We ask ourselves, then, how important is our father to our home? If our mother is what makes our home shine, and she is [...]

On January 6, 2017 Michelle Obama gave her last comments a First Lady. For a long time, she has served by President Barack Obama and she is currently confronting her goodbye. She studied sociology and African American Studies at [...]

These day people in the common era don’t seem to realize what the word hero means. This is due to the amount of celebrities and sport athletes that are famous. We are not realizing not every famous person is always a hero and [...]

In this digital era of comic books and fantasy movies, there are many amazing superheroes characters especially in marvel movies. Each superheroes has their own unique personalities. My favorite superheroes are Ironman and [...]

Not many people know who he is, yet we use what he made almost every day: Google, one of the most used websites on the internet. Google is a search engine which, 'organizes the world's information and makes it universally [...]

Carolus Linnaeus was born to Christina Brodersonia and Nils Linnaeus, a church minister and botanist. Linnaeus was born in Rashult, Sweden on May 23, 1707. Growing up, Linnaeus was educated in the subjects of latin, religion and [...]

Get Your Personalized Essay in 3 Hours or Less!

We use cookies to personalyze your web-site experience. By continuing we’ll assume you board with our cookie policy .

Instructions Followed To The Letter
Deadlines Met At Every Stage
Unique And Plagiarism Free

IMAGES

Role Model Essay
Awesome Role Model Essay ~ Thatsnotus
Reflection Essay: Essays on role models
role model essay my role model in life essay public speaking speech essay help write my research
essay role model
My role model essay introduction in 2021

VIDEO

Role Models 15
My father a role model English essay
Essay on Women Education//Importance of Women Education Essay//Advantage of Women Education Essay
Meri role model essay in hindi #shorts #youtubeshorts #trending #subscribe #calligraphy #essay
Paragraph on my role model gallantry award winner and the values I have learnt from his life are
Role classroom in nation building essay explained by Alevels Academy Islamabad Barrister Amna

COMMENTS

Essays About Role Models: Top 5 Examples and 7 Prompts
1. Best Role Model in Life by Prasanna. "But for me, a role model is someone who motivates me and influences me to be a good human and positive person in life, who gives me a supporting hand to walk and run to achieve my dreams and goals. Role model is not someone who is full of ego, attitude or full of selfishness.".
Role Model Essay
500 Words Essay On Role Model. A person whose behaviour, example, or accomplishment is or can be emulated by others, especially by younger people is how the term role model is defined. A positive, composed, and self-assured person is a wonderful role model. You don't want someone who attempts to bring you down.
Role Models Essay: Most Exciting Examples and Topics Ideas
A young scientist might see Marie Curie as a role model, or a young child might follow the example of a parent, teacher, or older sibling. Even adult might have a role model they aspire to be like. In English Composition, you may be asked by your professor to write an essay of your role model topics and why you follow their examples.
83 Role Model Topic Ideas & Essay Samples
A role model essay conclusion should include all of your arguments and the main points of the paper. Make sure that you use correct grammar and sentence structures. Check it several times to make sure that there are no mistakes in it. Pay attention to the spelling of words, too. If you are not sure that the grammar is correct, ask your peers to ...
My Role Model Essay: 500+ Words Essay in English for Students
We have provided 10 lines on My Role Model in English. In your essays and paragraphs for your exam and the school competition, you can include these lines. You can learn more about this subject from these lines. Everyone has a role model to whom they aspire. My inspiration is Mahatma Gandhi, a well-known freedom fighter.
Role Model Essay for Students and Children
500+ Words Essay on Role Model. Role Model Essay - There are many people in the world who blindly follow anyone. Also, they admire and love that person no matter what he does. A role model is a person who inspires you to be like him. Moreover, a role model is the person you love and want to be like him. Besides this, the role model can be ...
My Role Model Essay: A Few Ways To Elaborate On The Subject
Describe and illustrate specific qualities. My role model essays should be accurate. Avoid relying on vague language that merely depicts your role model as "brilliant" or "compassionate.". Instead, aim for a more penetrating portrait that captures your role model's complexities. Moreover, just because you're not writing about ...
Who Is Your Role Model: My Father's Influence on My Life: [Essay
Funny Hook Examples for Role Models Essay. The "Superhero" Next Door: While most people have superheroes on posters, my role model can be found in the house next door, armed with a lawnmower and an uncanny ability to grill the perfect burger. Move over, Batman; my neighbor is the true hero of the neighborhood! When Life Gives You Lemons, Call My Role Model: If life ever decides to bombard ...
Role Models Essay Examples
The purpose of such essay examples is to highlight the positive impact that role models have on personal development, education, and societal progress. These essays celebrate the qualities that make role models exceptional and encourage readers to reflect on the attributes they admire and aspire to emulate. 1. Inspiration and Empowerment.
How Role Models Influence Behaviors
Introduction. A role is a behavior or attention to duty expected of an individual to which he or she may adhere to another. A role model is a character implementing a certain role in society for others to copy. This is to say a role model is used as an example by other actors in the society. They can be teachers in various institutions whose ...
Free Essays on Role Models, Examples, Topics, Outlines
Role models essays are generally about people who we respect and want to resemble. Our role models essay samples feature many essays about remarkable people. Feel free to check out these samples of essays on role models, as they might help enrich your own essay.
Characteristics of a Good Role Model
In conclusion, a good role model possesses a unique set of characteristics that make them admirable and influential. Integrity, resilience, empathy, leadership, and open-mindedness are some of the key traits that define a good role model. By embodying these qualities, a good role model sets a positive example for others to follow and inspires ...
Why it's important to identify role models (essay)
That is incredibly valuable information because, as important as it is to identify role models, it's even more important to learn how they got to where they are today. Once you have identified several role models, you could contact them directly and ask for a brief conversation on a specific question. I melt whenever someone emails me and ...
Role Model Essays: Examples, Topics, & Outlines
Role Model Maya Angelou the. Words: 724. Length: 2 Pages. Document Type: Essay. Paper #: 97659167. Read Full Paper . Additionally she has time to lecturer and mentor students. This approach to one's career is very inspiring. She is doing what she wants to do.
Descriptive Essay: My Role Models
Descriptive Essay: My Role Models. Many people look to someone to serve as a model or motivator for their life. They want someone to look up to, who has good values and a sense of honor and integrity. Most people choose one person to act as their 'guiding light' for their life, but I am slightly different. I have chosen several.
Role Model Essay Examples
Thus, you can use role model essay samples for free as a writing guide or reference material for your academic paper. Our writers ensure each sample is 100% original and error-free. Manual for Essay Writing. Learners who use our samples get an effective guide on essay structures and content organization. You can search for relevant role models ...
My Role Model Essay
100 Words Essay on My Role Model. The one who gave birth to me is my role model. The amount of sacrifice and pain she had borne is unpayable. I look upon her as the strongest woman and someone who is a statue of morality. I always try to follow her guidelines and ideals. My mother is a living goddess, who always works for the well-being of ...
Essay on My Role Model
250 Words Essay on My Role Model Introduction. Parents are the first role models for any child. In my life, my parents have played a pivotal role, shaping my character, values, and aspirations. Their influence is not confined to my early years but continues to guide me in my adult life.
Emma Watson as My Role Model: [Essay Example], 671 words
As seen in many movies 'Emma Watson' is best known as Hermione Granger from the Harry Potter film series since 2001 (Wittmer, C. (2017). Also, one of the best actresses and she has won many awards since a young age from many award shows. Also, Emma has stunning performance and role in the remake of 'Beauty and the Beast' in Disney's ...
Why My Mom is My Role Model [Free Essay Sample], 511 words
In conclusion, the phrase "my mom is my role model" encapsulates the profound impact my mother has had on shaping my. values, character, and outlook on life. Her strength, compassion, empathy, and unwavering commitment to leading. by example have made her an exceptional role model. Through her actions and guidance, she has instilled in me a.
Artificial intelligence-driven radiomics study in cancer: the role of
Radiomics-based tumor staging. Some studies have focused on developing radiomics-based models for tumor staging as shown in Table 2 [17,18,19,20], which classifies the severity of a malignancy according to the size of the primary tumor and the extent of its spread throughout the body.Traditionally, tumor staging information is obtained by performing imaging examinations and pathological ...
My Sister is My Role Model [Admission Essay Example]
Published: Dec 27, 2022. In this essay, I will explore why my sister is my role model. Growing up, I admired her and aspired to follow in her footsteps. However, our relationship evolved when we moved to the United States, and I began to see her as not just a role model but also a close friend. Her college essay about family brought us even ...
The role of managed services in the transformation agenda
The role of managed services in the transformation agenda. Navigating the ongoing journey with flexible operating models. In our era of compound volatility, it's not enough to just transform. Many companies are doing so continually, with flexible operating models that keep evolving amid constant change. One way to create those operating ...
My Hero Speeches: an Exploration of The Importance of Role Models
We often find ourselves seeking guidance and inspiration in our day-to-day lives. Whether it be pursuing our academic goals, finding a career path, or simply navigating through personal challenges, we all turn to someone we admire and look up to during these times.These individuals, commonly known as our heroes, serve as powerful motivators and sources of support in our lives.
My Mother Is My Role Model: [Essay Example], 398 words
My role model is my mother. She is my greatest hero, and I admire her so much. Before her, my entire family was living in the countryside. However, she moved out to the city when she was only 18, starting with nothing. She and my father started a clothes shop with the money they earned. Even when she got pregnant, my mother continued to work ...

Essays About Role Models: Top 5 Examples and 7 Prompts

1. Best Role Model in Life by Prasanna

7 Writing Prompts for Essays About Role Models

Role Model Essay for Students and Children

My Role Model

His Achievement

His Personal Life

Qualities of a Role Model

How to be a Role Model for Others?

My Other Role Model

Customize your course in 30 seconds

Leave a Reply Cancel reply

Download the App

My Role Model Essay: A Few Ways to Elaborate on The Subject

Popular Posts

Related Posts

How Role Models Influence Behaviors Essay

Essays on Role Models

Related topic to Role Models

Creative samples from the experts

Descriptive Essay: My Role Models

Follow Us on Social Media

Get more free essays

Send via email

Contributors Bio

Find more useful services for students

Products & Resources

Colleges & Courses

Predictors & E-Books

Animation Courses

Top Courses & Careers

Quick Links

Diploma Colleges

Upcoming Events

Other Exams

Top Countries

Student Visas

Engineering Preparation

Medical Preparation

Online Courses

Top Streams

Specializations

Top Providers

My Role Model Essay

200 Words Essay on My Role Model

Qualities of My Role Model

Inspiring Quotes by Ratan Tata

Reasons Why He is a Role Model

Explore Career Options (By Industry)

Data Administrator

Bio Medical Engineer

Ethical Hacker

Data Analyst

Geothermal Engineer

Database Architect

Remote Sensing Technician

Budget Analyst

Underwriter

Finance Executive

Operations Manager

Stock Analyst

Welding Engineer

Transportation Planner

Environmental Engineer

Safety Manager

Conservation Architect

Structural Engineer

Highway Engineer

Field Surveyor

Orthotist and Prosthetist

Pathologist

Veterinary Doctor

Audiologist

Video Game Designer

Radio Jockey

Choreographer

Social Media Manager

Photographer

Copy Writer

Corporate Executive