In a study, AI reads cancer survival risk from a face photo

A single photograph, one face, and an estimate of how old the body behind it really is. That is what a new pair of AI tools from Mass General Brigham produces, according to a study published in The Lancet Digital Health. The tools were built by researchers at the Boston hospital network and rest on a shared base model called FAHR-Face, short for Foundation Artificial Intelligence for Health Recognition.

The key idea is a foundation model: a system first trained to learn a general picture of the human face from more than 40 million face images, and then adapted to specific health questions. The team, as co-senior author Hugo Aerts put it, did not train a separate model from scratch for each task. From that base, they built FAHR-FaceAge, which estimates biological age with greater accuracy than earlier age-estimation algorithms, and FAHR-FaceSurvival, which predicts survival outcomes in patients with cancer from a face photograph. The two tools captured different information: their predictions were only weakly correlated, and combining them further improved the fit of a survival model in an external validation cohort of cancer patients. The gap between a patient's FAHR-FaceAge and their chronological age was independently linked to survival.

The tools do not look at the same parts of the face. FAHR-FaceAge focuses on the nasolabial folds, the forehead just above the eye sockets and the temples. FAHR-FaceSurvival examines the area below the eyes and around the bridge of the nose. The study also found that lifestyle shows up on the face: heavy alcohol use, drug use and smoking were associated with looking biologically older, and smoking produced the strongest pattern, with current smokers looking older than former smokers, who looked older than people who had never smoked. The authors tested robustness as well. FAHR-FaceAge stayed consistent across pose, lighting, facial expression and makeup, and it was stable across most plastic surgery procedures tested, with only a small shift after rhinoplasty.

For oncology teams, the gain would be a second, quantitative signal before they recommend an intensive treatment. Mak notes that the population is ageing and that physicians need better ways to estimate physical reserve and frailty before choosing radiation, surgery or systemic therapy. The people who could benefit are cancer patients facing those decisions. But the timeline is still research: these are validation results from one study, in cancer patients, and they have not been shown to work as routine clinical tools. Mak's stated goal is to make face-based assessment "more quantitative, consistent and scalable" so that it supports personalized treatment decisions rather than replacing physician judgment. The team is also running the FaceAge Research Study, an online study approved by an institutional review board and open to the public, in which participants upload a photograph, receive their own FaceAge readout and contribute data to help validate the measure across larger and more diverse populations.