Scientists develop AI tool to predict biological age and cancer survival using just a selfie
Now, a new tool powered by artificial intelligence (AI) may be able to tell how quickly you're ageing, using only a selfie – not to insult or flatter you, but to assess your health.
For a new study published in The Lancet Digital Health journal, researchers used photos of nearly 59,000 faces to train an AI model to estimate people's biological ages, or their age based on their cellular health rather than their birth date.
Related
Having a poor socioeconomic background could speed up biological ageing, new study finds
Then they took the model, called "FaceAge," to about 6,200 cancer patients.
On average, cancer patients looked about five years older than their actual ages, and they tended to have higher FaceAge readings than people without cancer, the study found.
Notably, the model also helped doctors make better predictions about the short-term life expectancies of cancer patients receiving palliative care. Only the best physicians' predictions compared to FaceAge alone on accuracy.
"How old someone looks compared to their chronological age really matters," Hugo Aerts, one of the study's authors and director of the AI in medicine programme at Mass General Brigham in the US, said in a statement.
Related
Study on 'fragile' AI predictive models provides 'cautionary tale' about use in medicine
The researchers said that eventually, the tool could help doctors and cancer patients make decisions about end-of-life care – but that it could also be used to address a host of other health issues.
Dr Ray Mak, one of the study authors and a cancer physician at Mass General Brigham, said FaceAge could someday be used as an "early detection system" for poor health.
"As we increasingly think of different chronic diseases as diseases of ageing, it becomes even more important to be able to accurately predict an individual's ageing trajectory," Mak said in a statement.
Related
Is your heart ageing faster than you are? UK scientists develop tool to track its 'functional age'
The tool has some limitations. It was primarily trained on white people, and it's not clear how factors that affect people's appearances – like lighting or make-up – could shape the results.
The researchers are now expanding their work to include more hospitals and cancer patients at different stages of the disease, as well as testing FaceAge's accuracy against datasets with plastic surgery and make-up.
Actually seeing a tool like FaceAge used in the doctor's office is a long way away.
But Mak said it "opens the door to a whole new realm of biomarker discovery from photographs".
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
2 days ago
- Yahoo
The app will see you now: New AI scans faces to predict diseases, disorders, and early death
Facial recognition apps detect pain in dementia patients, trauma in children, and diagnose infections. I tried Harvard's FaceAge app using photos to estimate biological age and, in turn, overall health. This article is part of "Transforming Treatments," a series on medical innovations that save time, money, or discomfort. I look like I'm about 28. Or maybe 38. That's according to Harvard's "FaceAge" algorithm, which uses photos to determine a person's supposed biological age — meant as a quick proxy for wellness. This app is one of several new efforts to turn selfies into diagnostic tools. There's one for diagnosing nasal congestion, another for seasonal allergies, and a few safe driving apps that watch your face for signs of drowsiness. Some face scanners measure pain, illness, or signs of autism. One aims to track PTSD in kids to spare them from having to talk about traumatic issues over and over again. Since 2022, facial recognition for the clinic has blossomed, alongside rapid advancements in artificial intelligence and chipmaking. This year, new face technologies promising to diagnose diseases earlier, treat patients better, and ostensibly predict early death are taking off. "It's a medical biomarker, not just a gimmick," said FaceAge creator and radiologist Dr. Raymond Mak, who's leading the team at Harvard and Mass General Brigham developing facial recognition technology that Business Insider recently tested. Ethics experts worry about what we're barreling into, without better understanding exactly what this tech is measuring, or why. "AI is entering these spaces fast," Malihe Alikhani, an assistant professor of machine learning at Northeastern University, told Business Insider. "It's about making sure that it is safe and beneficial." Your face is a reflection of your health Our faces say a lot about our physical, mental, and biological health. While this is new territory for computers, humans have read faces to make quick judgments for thousands of years. Research suggests we developed a third type of cone in our eyeballs about 30 million years ago, specifically to scan each other's faces for signs of health or sickness. That cone allows us to read faces in shades of red and green. "People look at rosy cheeks and they see that as a sign of health. When we want to draw a face that's sick, we'd make it green," Professor Brad Duchaine, a neuroscientist who studies facial perception at Dartmouth, told Business Insider. It's true: A flush can indicate good blood flow, or high levels of carotenoids in the skin from fruits and veggies we eat. Dr. Bahman Guyuron, a plastic surgeon in Cleveland, studied identical twins with different lifestyles to see how factors like smoking and stress impact their faces. Consistently, the twin with more stress and more toxins in their bodies looked several years older. Sagging skin and wrinkles can reflect poorer internal health, with lower collagen production and higher levels of stress hormones. Conversely, studies show that superaging centenarians — whose organs and cells are working unbelievably efficiently — look, on average, about 27 years younger than they are. I tried a face scanning app One of the first medical applications of face-reading tech was Face2Gene, an app first released in 2014 that helps doctors diagnose genetic conditions. Studies suggest Face2Gene is better than human doctors at extracting information from a person's face and then linking those features to a specific genetic issue. The Australian app PainChek has tracked the pain of nursing home patients since 2017. It is mostly used for dementia patients who may not be able to verbalize pain. In a recent announcement, the company said it is awaiting FDA approval and could be cleared by September 2025. I wanted to try one of these apps for myself. Since I write a lot about aging, I decided to try FaceAge, Harvard's new app that ostensibly measures your biological age. It is not yet available for public use; it is being used as a research tool for now. The ultimate goal, researchers say, would be to use selfies to do better diagnostic work. FaceAge could one day improve cancer treatments by tailoring them to a patient's unique biology and health status, or maybe even help flag other health issues before they happen. The FaceAge algorithm pays attention to two main areas of the face: the nasal labial folds, from the nose to the lips, and the temples between the eyes and ears. The idea is to spot premature or accelerated signs of aging that could be a red flag for internal problems. "If your face age is accelerating quicker than your chronological age, it's a very poor prognostic sign," Mak, one of the developers behind FaceAge, told Business Insider. I submitted four photos to the app. In the darkest, blurriest photo I provided, the app thought I was 27.9 years old — a little more than a decade below my actual age. The picture I took with no makeup on, and my face thrust out into the bright mid-day sunshine, gave me the oldest FaceAge, even though all of these pictures were taken within the past year. One selfie taken in the dark of winter and another on a cloudy day ended up somewhere in the middle, making me look young-ish. Humans (and face-scanning apps) use the proliferation of lines, sharp edges, and more details to assess someone's age. That's why people look younger in blurry photos — or with surgery or makeup to smooth over their wrinkles. In "a really, really, really blurry photograph of a face, what you've done is you've stripped out all of the high spatial frequency information," neuroscientist Bevil Conway, a senior investigator at the National Institutes of Health, told Business Insider. Direct light, like a ring light, can help mask old age. The midday sun, coming down on my face from above, had the opposite effect. So, what did I learn from my experiment? FaceAge told me I'm looking great (and young!) and should keep up my healthy habits. Still, its assessments were all over the place. Face Age is confident each time you run it, but that confidence masks the fact that it can't really tell how well I'm aging over time. Is my body 10 years younger than me, or just one? While I do eat a relatively healthy diet and exercise regularly, I'm curious how much the differences in lighting affected my results. The ethics are blurry Even if it's something as seemingly innocuous as measuring your age, bringing AI into the doctor's office is fraught with ethical conundrums. "We have been through a few years now of companies coming up with these systems, selling them to hospitals, selling them to doctors, selling them to border protection, and then after a while they're like, 'oops,'" Alikhani, the AI ethics expert, said. Readers may remember the uproar over the highly controversial Stanford study that developed "gaydar" AI in 2017. The app purported to spot someone's sexual orientation. Critics said it was just picking up on social and environmental cues that had nothing to do with sexuality, like makeup, facial hair, and sun exposure. Another team of researchers from Shanghai Jiao Tong University developed an algorithm that promised to identify criminals and terrorists, or people with law-breaking tendencies. These efforts feel uncomfortably close to the pseudoscientific practice of physiognomy, a deeply flawed practice that's been used for centuries to justify racism and bigotry, Alikhani said. Facial expression is highly context-dependent, varying not only based on a person's gender and culture, but also by the individual and the moment, she said. "Better healthcare involves patients more in the decision-making," Alikhani said. "What are we doing if we're putting that in the hands of AI?" Read more from the Transforming Treatments series: In an era of quiet glow-ups, no-prep veneers are the new 'it' cosmetic procedure Colonoscopies are no fun. These at-home colon cancer screenings offer a shortcut. Skin tightening is getting more advanced — and less painful. Here are the new techniques replacing facelifts. Read the original article on Business Insider
Business Insider
09-08-2025
- Business Insider
The app will see you now: New AI scans faces to predict diseases, disorders, and early death
I look like I'm about 28. Or maybe 38. That's according to Harvard's "FaceAge" algorithm, which uses photos to determine a person's supposed biological age — meant as a quick proxy for wellness. This app is one of several new efforts to turn selfies into diagnostic tools. There's one for diagnosing nasal congestion, another for seasonal allergies, and a few safe driving apps that watch your face for signs of drowsiness. Some face scanners measure pain, illness, or signs of autism. One aims to track PTSD in kids to spare them from having to talk about traumatic issues over and over again. Since 2022, facial recognition for the clinic has blossomed, alongside rapid advancements in artificial intelligence and chipmaking. This year, new face technologies promising to diagnose diseases earlier, treat patients better, and ostensibly predict early death are taking off. "It's a medical biomarker, not just a gimmick," said FaceAge creator and radiologist Dr. Raymond Mak, who's leading the team at Harvard Medical School developing facial recognition technology that Business Insider recently tested. Ethics experts worry about what we're barreling into, without better understanding exactly what this tech is measuring, or why. "AI is entering these spaces fast," Malihe Alikhani, an assistant professor of machine learning at Northeastern University, told Business Insider. "It's about making sure that it is safe and beneficial." Your face is a reflection of your health Our faces say a lot about our physical, mental, and biological health. While this is new territory for computers, humans have read faces to make quick judgments for thousands of years. Research suggests we developed a third type of cone in our eyeballs about 30 million years ago, specifically to scan each other's faces for signs of health or sickness. That cone allows us to read faces in shades of red and green. "People look at rosy cheeks and they see that as a sign of health. When we want to draw a face that's sick, we'd make it green," Professor Brad Duchaine, a neuroscientist who studies facial perception at Dartmouth, told Business Insider. It's true: A flush can indicate good blood flow, or high levels of carotenoids in the skin from fruits and veggies we eat. Dr. Bahman Guyuron, a plastic surgeon in Cleveland, studied identical twins with different lifestyles to see how factors like smoking and stress impact their faces. Consistently, the twin with more stress and more toxins in their bodies looked several years older. Sagging skin and wrinkles can reflect poorer internal health, with lower collagen production and higher levels of stress hormones. Conversely, studies show that superaging centenarians — whose organs and cells are working unbelievably efficiently — look, on average, about 27 years younger than they are. I tried a face scanning app One of the first medical applications of face-reading tech was Face2Gene, an app first released in 2014 that helps doctors diagnose genetic conditions. Studies suggest Face2Gene is better than human doctors at extracting information from a person's face and then linking those features to a specific genetic issue. The Australian app PainChek has tracked the pain of nursing home patients since 2017. It is mostly used for dementia patients who may not be able to verbalize pain. In a recent announcement, the company said it is awaiting FDA approval and could be cleared by September 2025. I wanted to try one of these apps for myself. Since I write a lot about aging, I decided to try FaceAge, Harvard's new app that ostensibly measures your biological age. It is not yet available for public use; it is being used as a research tool for now. The ultimate goal, researchers say, would be to use selfies to do better diagnostic work. FaceAge could one day improve cancer treatments by tailoring them to a patient's unique biology and health status, or maybe even help flag other health issues before they happen. The FaceAge algorithm pays attention to two main areas of the face: the nasal labial folds, from the nose to the lips, and the temples between the eyes and ears. The idea is to spot premature or accelerated signs of aging that could be a red flag for internal problems. "If your face age is accelerating quicker than your chronological age, it's a very poor prognostic sign," Mak, one of the developers behind FaceAge, told Business Insider. I submitted four photos to the app. In the darkest, blurriest photo I provided, the app thought I was 27.9 years old — a little more than a decade below my actual age. The picture I took with no makeup on, and my face thrust out into the bright mid-day sunshine, gave me the oldest FaceAge, even though all of these pictures were taken within the past year. One selfie taken in the dark of winter and another on a cloudy day ended up somewhere in the middle, making me look young-ish. Humans (and face-scanning apps) use the proliferation of lines, sharp edges, and more details to assess someone's age. That's why people look younger in blurry photos — or with surgery or makeup to smooth over their wrinkles. In "a really, really, really blurry photograph of a face, what you've done is you've stripped out all of the high spatial frequency information," neuroscientist Bevil Conway, a senior investigator at the National Institutes of Health, told Business Insider. Direct light, like a ring light, can help mask old age. The midday sun, coming down on my face from above, had the opposite effect. So, what did I learn from my experiment? FaceAge told me I'm looking great (and young!) and should keep up my healthy habits. Still, its assessments were all over the place. Face Age is confident each time you run it, but that confidence masks the fact that it can't really tell how well I'm aging over time. Is my body 10 years younger than me, or just one? While I do eat a relatively healthy diet and exercise regularly, I'm curious how much the differences in lighting affected my results. The ethics are blurry Even if it's something as seemingly innocuous as measuring your age, bringing AI into the doctor's office is fraught with ethical conundrums. "We have been through a few years now of companies coming up with these systems, selling them to hospitals, selling them to doctors, selling them to border protection, and then after a while they're like, 'oops,'" Alikhani, the AI ethics expert, said. Readers may remember the uproar over the highly controversial Stanford study that developed "gaydar" AI in 2017. The app purported to spot someone's sexual orientation. Critics said it was just picking up on social and environmental cues that had nothing to do with sexuality, like makeup, facial hair, and sun exposure. Another team of researchers from Shanghai Jiao Tong University developed an algorithm that promised to identify criminals and terrorists, or people with law-breaking tendencies. These efforts feel uncomfortably close to the pseudoscientific practice of physiognomy, a deeply flawed practice that's been used for centuries to justify racism and bigotry, Alikhani said. Facial expression is highly context-dependent, varying not only based on a person's gender and culture, but also by the individual and the moment, she said. "Better healthcare involves patients more in the decision-making," Alikhani said. "What are we doing if we're putting that in the hands of AI?"
Medscape
30-07-2025
- Medscape
Can 3 Questions Flag Depression? PHQ-3 Tested
A brief, 3-question version of the widely used Patient Health Questionnaire-9 (PHQ-9) may provide a faster, yet still accurate, way for physicians to screen for depression, new research suggests. In a study of more than 96,000 US adult participants, the PHQ-3 had 98% sensitivity and 76% specificity for predicting at least moderate or greater depressive symptoms. Additionally, it was 'highly correlated' with the full PHQ-9 and noninferior to a recently validated 4-item version for effectiveness — and it yielded a sensitivity of more than 90% across all ages, races, and ethnicities. Lead investigator Roy H. Perlis, MD, vice chair of research in the Department of Psychiatry at Mass General Brigham in Boston, said the team set out to identify the shortest version of the PHQ-9 that could still effectively screen for depression — and found that the PHQ-3 hit the 'sweet spot.' 'We let the data tell us what the constellation of symptoms are that best predict overall severity and how well they do at identifying depression. It's a way we might be able to get the maximum amount of information in the shortest amount of time,' Perlis told Medscape Medical News. However, he noted that he's not yet ready to endorse the PHQ-3 for use in clinical practice. 'I want to see further studies in clinical settings. But I think we made a good start towards showing it did well in a big, general sample of US adults,' Perlis said. The findings were published online on July 21 in JAMA Network Open . Shorter Attention Spans 'The PHQ-9 was not intended as a tool to diagnose depression. It's just a way to start the conversation; and although it's not perfect, it is useful,' Perlis noted. The original patient-reported PHQ screens for depression, anxiety, eating disorders, alcohol misuse, and somatization. Its 9-item depression scale (PHQ-9) became widely used to identify depression along with related symptoms such as fatigue and sleep disturbances. However, the length of the questionnaire has become burdensome for some patients, particularly those who prefer using phone apps over printed forms. 'Faced with a long list of survey questions, some individuals may be tempted to speed through or to not respond at all,' the investigators wrote. At the recent American Psychiatric Association (APA) annual meeting, clinicians in the audience at several sessions mentioned that their patients have been complaining that the PHQ-9 is too long and that they prefer data to be delivered in smaller 'chunks.' Perlis said that attitude was a big motivator for the study. 'We're in an era where we have shorter attention spans and people want things very quickly,' he said. 'Personally, I would always rather have the PHQ-9. The question we were trying to answer was: Can we ask fewer questions if we don't have the time or the space to ask about nine [items] and do almost as well?' Perlis said. The researchers assessed data from four waves of an online survey conducted from November 2023 to July 2024. In the first wave, they identified the optimal questionnaire items to be included. The four waves had a total of 96,234 participants (57% women; mean age, 47.3 years). Of these, 68% were White, 13% were Black, 10% were Hispanic or Latino, 5% were Asian, and 4% were classified as 'other.' In the full patient population, 26% had moderate or greater depressive symptoms, as measured by a PHQ-9 score ≥ 10. Follow-up Is Key After examining shortened versions of the PHQ-9 that ranged from including just one item up to eight items, the PHQ-3 with items 1 (interest), 2 (depressed mood), and 6 (self-esteem or failure) was deemed to be the 'optimal' version. It had a sensitivity of 0.98 (95% CI, 0.97-0.98) and a specificity of 0.76 (95% CI, 0.75-0.76) for moderate or greater depressive symptoms. Across all subgroups except participants aged 65 years or older, the sensitivity for the PHQ-3 was greater than 0.94. For that subgroup, it was 0.93. The PHQ-3 was also noninferior to previously reported sensitivity and specificity of the PHQ-Depression-4 in the whole study group and in all subgroups. The area under the receiver operating characteristic curve (AUROC) for predicting moderate or greater depressive symptoms was 0.83 for the PHQ-3. The AUROC for the PHQ-9 was 0.84. 'While a shortened scale cannot capture the full range of the PHQ-9, it may facilitate more widespread and efficient investigation of psychiatric symptoms in general population samples when participant burden and/or data collection expense must be minimized,' the investigators wrote. Overall, Perlis said the PHQ-3 could be a possible first-step screening tool, with more questions added as needed. 'What's most important is that people are screening for depression and, if someone screens positive, that they're following up with a more comprehensive evaluation and offering treatment if it's indicated,' he said. 'It's really about figuring out who needs follow-up.' Menu of Options? In an accompanying editorial, Kurt Kroenke, MD, of Regenstrief Institute, Indianapolis, Indiana, noted that the study had 'numerous strengths,' including its large and diverse study population, as well as several limitations. First, the criteria for calculating sensitivity and specificity did not include a structured psychiatric interview — the standard used in previous evaluations of the 2- and 4-item PHQs. In addition, because the researchers drew from a general population sample, future studies should focus on actual patients in both primary care and psychiatric settings, Kroenke suggested. He noted that, in addition to the various shortened PHQs, the 2-item and 4-item versions of the Patient-Reported Outcomes Measurement Information System (PROMIS) have also demonstrated benefit. 'Clinicians and researchers looking for ultrashort depression measures now have a menu from which to choose, which is good since one size may not fit all purposes,' Kroenke wrote. 'Expanded use of brief screeners to increase detection of depression has the potential to decrease the burden of the most prevalent mental disorder worldwide,' he added.
