.png "Global Insight")
What if a single photograph of your face could help predict how long a cancer patient might survive? Researchers at Mass General Brigham in Massachusetts say their AI can do exactly that — and their study, published Monday in The Lancet Digital Health, is being called the most comprehensive demonstration yet of the idea.
This article is general information only and is not medical advice.
Here is how the system works, what it actually found, and the important reasons doctors say it is not ready for your phone yet.
Two tools, one face
The team built a foundation model trained on more than 40 million face images, then tuned it on photographs of more than 30,000 cancer patients. From that base, they developed two complementary tools. FAHR-FaceAge estimates a person's biological age — how old they look rather than their chronological age — and links it to survival. FAHR-FaceSurvival predicts survival outcomes directly from the photograph.
The results, according to the Straits Times, were striking. In a test group of 3,820 cancer patients, more than half of those given the highest risk score — 64 out of 122 — died within 90 days of the photograph being taken. In a separate group of more than 38,000 cancer patients, those whose biological age came out five or more years older than their real age faced a 21 percent higher risk of death; those who looked five-plus years younger had a 17 percent lower risk.
What the AI is actually "seeing"
Using a technique called attention mapping, the researchers traced where the algorithm looks. For age estimates, it focused on the nasolabial folds (the creases from nose to mouth corners), the forehead and the temples. For survival risk, it zeroed in on the bridge of the nose and the areas below the eyes. Why those regions correlate with mortality is still unclear.
The tool held steady across variations in pose, lighting, expression and makeup, and worked across ages, races, genders and cancer types. "To our knowledge, this is the first demonstration that an AI model can directly predict cancer survival from a photograph," said Hugo Aerts, the study's co-senior author. Corresponding author Raymond Mak, a radiation oncologist, said the goal is to help doctors gauge frailty before recommending grueling treatments — "rather than replace physician judgment."
The caveats
As Scientific American notes, independent experts urge caution: Miroslaw Bober, a machine-learning professor at the University of Surrey, said the tool is not ready for clinical use. A high risk score is a statistical signal, not a diagnosis — and telling a patient an algorithm scored them likely to die within 90 days raises ethical questions no study has answered yet.
Still, the direction is clear. AI is moving from reading scans to reading people — from reviving frozen organs to decoding faces. The selfie, it turns out, may be a medical instrument. For more on the frontiers of medicine, see our coverage of Pennsylvania's congenital measles alert.
Would you want an AI to estimate your health from a photo — or is some knowledge better left unknown? Follow Global Insight for the science that shapes our future.