Photo: Thirdman/Pexels
Call it "selfie science," perhaps. Researchers are finding that ubiquitous facial images could be used to train artificial intelligence for potential deployment as a clinical decision support tool – assessing a patient's biological age when deciding on treatment.
An upcoming presentation scheduled for March at the 2026 HIMSS Global Health Conference & Exposition in Las Vegas will provide an overview and dive into results from clinical trials of the deep learning algorithm called FaceAge, which analyzes facial images to predict a person’s biological age and survival factors.
The thinking behind this AI development is that facial analysis may offer more objectivity than a doctor's visual intuition alone and enhance doctors' qualitative assessments, says Dr. Raymond Mak, a radiation oncologist with Brigham and Women’s Hospital in Boston and researcher with Mass General Brigham's AI in Medicine Program.
"What we think is that there is underlying biology, captured in your facial tissues – the skin, the muscles, the blood flow," he told Healthcare IT News on Thursday.
Testing doctors' predictions
Healthcare has been measuring patient biological age through genomics, blood-based tests and imaging, and researchers who are testing how AI and simple face photographs could help say the technology holds promise.
"We demonstrated a proof of principle that's a novel prognostic factor in cancer patients, and it can be illustrated as a clinical use case," said Mak.
Doctors generally consider age an important factor for making difficult medical decisions, he noted. And some doctors are "really good" at assessing their patients' biological ages."
"We had a couple of doctors perform at levels where they're right about 80% of the time," he said. "They're picking up on something – some visual cues."
Over the last year, the researchers trained FaceAge on more than 58,000 images of healthy individuals and on 6,196 cancer patients.
Then, in comparison studies of palliative care patient data and known outcomes, the algorithm outperformed clinicians in predicting short-term life expectancy (less than or greater than six months until death), he explained.
"When we asked doctors to guess that, just based on the photograph – the eyeball tests – they're only slightly better than a coin flip on average," Mak said.
Doctor performance improved by about 10% when standard clinical information was provided to them, but when the researchers gave the doctors FaceAge data, "that's when their performance improved" to predict correctly seven or eight times out of 10.
The research report, which Mak said will be published soon in Nature, revealed that cancer patients typically had a FaceAge five years older than their actual age, and those with older facial analysis had worse survival rates.
Potential use cases and criticisms
FaceAge and a second algorithm in development, FaceSurvival, capture "different domains of somebody's facial health, such that when we put them into prognostic models, they independently – together, but independently – provide additional prognostic power," Mak said.
While facial images are widely available and impose little cost to obtain, he said future research could also train the deep learning models on hand data, or potentially medical imaging like CT scans – as long as a patient's images over time are available for analysis.
However, these tools already have their critics, according to Mak.
While the technology could offer a non-invasive way to predict biological health, it raises ethical and clinical concerns, ranging from the perception that an AI prediction could be used to deny life-saving treatments.
But the benefits are intriguing, not just for doctors who want to improve their qualitative age assessments, but for patients.
"From the very beginning of medical school, we're taught to look at a patient and document whether the patient looks older than stated age, or the patient looks younger than stated age," Mak explained.
"These are all ways that doctors – through clinical gestalt – assess the patient's baseline health to make all kinds of big decisions," such as whether a patient can withstand major heart surgery or a more intense cancer treatment.
In some cases, Mak has seen patients enter chemotherapy, and within weeks, it looks as if "they aged 20 years," he said. "But that's not everybody."
The data could be prescient in a way that some oncology patients with genetic markers, for example, might hope would help reduce the uncertainties of their medical decisions.
Perhaps FaceAge and FaceSurvival could help doctors better answer the question, "When is this going to happen to me?" or help metastatic cancer patients decide whether or not to accept treatments that ultimately buy them less time to live than they would hope for and hold high risks of side effects.
Other patients may not want to know for psychological reasons or because they might be concerned that the data could be used to deny health plan coverage for treatments they want.
"That's not the goal; the goal is anywhere in medicine that we use chronological age to make a decision," said Mak. "FaceAge could help doctors improve their clinical gestalt."
While tests thus far showed the algorithm's efficacy despite face lifts or other plastic surgeries, the researchers are looking to test if weight loss, such as patients taking the weight loss drug semaglutide, and facial injury confuse the algorithm.
They're also collaborating with surgeons worldwide on future studies that explore FaceAge analysis applications in health assessments, such as in cardiovascular and neurological specialties.
"FaceAge: Using Artificial Intelligence to Decode Biological Age with a Selfie" is scheduled for Wednesday, March 11, from 2-2:30 p.m. in Level 5, Palazzo D in the Venetian, at HIMSS26 in Las Vegas.
Andrea Fox is senior editor of Healthcare IT News.
Email: afox@himss.org
Healthcare IT News is a HIMSS Media publication.
