AI predicts cancer survival from a single photo — looking 5 years older raises death risk by 21%

An AI model that predicts a cancer patient's likelihood of survival from a single facial photograph has been developed by researchers in the United States.

According to The Guardian and Mass General Brigham (MGB), a Harvard Medical School-affiliated hospital group, MGB researchers developed an AI model called FAHR-Face that estimates a person's biological age from a facial photo and directly predicts cancer survival. The findings were published Monday (local time) in the international journal The Lancet Digital Health.

The research team trained the base model on more than 40 million facial photographs spanning diverse ethnicities and age groups.

One component, FAHR-FaceAge, estimates apparent biological age — rather than chronological age — to gauge cancer survival prospects. A second component, FAHR-FaceSurvival, bypasses the biological-age estimate entirely and predicts survival probability directly from the facial image.

When the team evaluated the model's predictions, cancer patients in the United States who had undergone radiation treatment and whom FAHR-FaceAge judged to look five or more years older than their actual age faced a 21 percent higher risk of death over an average four-year follow-up period compared with other patients. Those judged to look five or more years younger than their actual age had a 17 percent lower risk of death.

Hugo Aerts, director of the MGB Artificial Intelligence in Medicine (AIM) program and a co-corresponding author of the study, told reporters that FAHR-FaceSurvival "advances facial health perception AI one step further by learning directly from facial photos of cancer patients," adding that it is "the first demonstration, to our knowledge, that an AI model can directly predict cancer survival from a simple photograph."

Raymond Mak, a radiation oncologist at the MGB Cancer Center, said the results were consistent despite everyday variations in photos such as posture, lighting, facial expression and makeup. "It was also stable after most cosmetic surgeries, with only slight changes observed after rhinoplasty," he said.

The researchers said, however, that broader use of the tool would require further research. The current results remain at the research stage and the tool is not yet ready for clinical use. MGB has filed a US patent application for the technology.

Earlier, the team had published a related AI model called FaceAge in the same journal in May 2025, which estimated biological age from facial photographs. In that study, a model trained on about 60,000 photos of healthy individuals was applied to more than 6,000 cancer patients, finding that cancer patients appeared, on average, about five years older than their actual age.