Mass General Brigham AI reads cancer survival from a face photo

A new Mass General Brigham study reports that an AI model called FAHR-FaceSurvival, given a single photograph of a cancer patient's face, flagged 122 patients as high-risk; 52.5% of them died within 90 days of the assessment, compared with about 2% of 975 patients the model placed in its lowest-risk group. The work, reported by CNN and published in The Lancet Digital Health, is being described by its authors as the first demonstration that survival in cancer can be predicted directly from a face photo.

FAHR-FaceSurvival sits on top of a broader foundation model, FAHR-Face, trained on more than 40 million facial images. A companion tool, FAHR-FaceAge, estimates biological age from the same input; among radiotherapy patients, those whose biological age came in more than 20 years above their chronological age had a 21% mortality rate across a 1,060-patient cohort, against roughly 6% among 4,195 patients whose face looked at least 10 years younger. The survival model focuses on the bridge of the nose and the area beneath the eyes, the age model on nasolabial folds, forehead and temples.

"FAHR-FaceSurvival advances the field of facial health recognition AI a step further by learning survival risk directly from facial photographs of patients with cancer," said Hugo Aerts, PhD, director of the Artificial Intelligence in Medicine program at Mass General Brigham and co-senior author on the paper, calling it "the first demonstration that an AI model can directly predict cancer survival from a simple photograph." Corresponding author Raymond Mak, a radiation oncologist at the Mass General Brigham Cancer Institute, has framed the tools as a supplement to existing prognostic workups rather than a replacement.

Outside experts urged caution. "This is an extremely compelling study that stems from a large data set," said Wilbur Lam of Emory University, while emphasising that prospective clinical validation across other hospitals and patient populations is still needed. The authors flag the same limits: the training cohort was majority white and Boston-area, and rhinoplasty was one of the few interventions that shifted the age estimate meaningfully.

The paper lands in an unusually busy stretch for the field; our healthcare AI tracker has logged 65 stories in the last 90 days alone. A public research portal at faceage.bwh.harvard.edu now lets volunteers submit their own photos, which the team says will help broaden future training data.