A new artificial intelligence model can look at a single photograph of a cancer patient's face and estimate how likely that person is to survive the next 90 days. Developed by researchers at Mass General Brigham, the tool identified high-risk patients who went on to die within three months at a rate of 52.5%, compared with roughly 2% among patients the model placed in its lowest-risk category.
A Model Built on Tens of Millions of Faces
The underlying system, called FAHR-Face, short for Foundation Artificial Intelligence for Health Recognition, was detailed in the peer-reviewed journal The Lancet Digital Health on Monday, according to Mass General Brigham. As reported by News12 New York, the model was trained using more than 40 million facial images before being fine-tuned on clinical data. According to research findings summarized by The ASCO Post, that fine-tuning drew on photographs and medical outcomes from 34,363 cancer patients, building on a foundation model pre-trained using masked autoencoders.
News12 New York reports that Hugo Aerts, one of the researchers involved, said the model directly predicts cancer survival from a simple photograph. The outlet also notes that Raymond Mak, a radiation oncologist involved in the work, said physicians need better ways to evaluate patients before recommending treatments such as radiation, surgery, and systemic therapy. Mak said the goal is to make face-photo-based health assessment more quantitative, consistent, and scalable, in order to support personalized treatment decisions.
Reading Age and Risk From Different Parts of the Face
The system is actually split into two tools that look at different facial landmarks. Per AI Weekly, the FAHR-FaceSurvival tool analyzes the bridge of the nose and the lower eye areas to assess mortality risk, while the FAHR-FaceAge tool evaluates nasolabial folds, forehead lines, and temples to estimate a patient's biological age. Among patients undergoing radiotherapy, those whose estimated biological age ran more than 20 years older than their actual age faced a 21% mortality rate, the outlet reports, compared with roughly 6% for patients who appeared at least 10 years younger than their chronological age.
Mass General Brigham's research also found that lifestyle factors such as heavy alcohol use, drug use, and smoking were significantly associated with an advanced biological age estimate during the facial analysis, according to the health system. That finding builds on earlier work: an April 2026 study published in Nature Communications established that tracking changes in a patient's biological age across sequential photos over time, a measure the researchers termed the Face Aging Rate, functioned as an active prognostic biomarker in its own right.
Built to Withstand Makeup, Lighting and Most Surgery
One practical question for any face-reading algorithm is whether it holds up outside a controlled lab setting. Sensitivity testing reported by Becker's Oncology showed the FAHR-FaceAge algorithm maintained predictive stability across variations in lighting, pose, facial expressions, and makeup, as well as most plastic surgery procedures, with only rhinoplasty producing minor shifts in results. News12 New York similarly reported that cosmetic changes, including facial surgery and Botox, did not meaningfully affect the model's performance.
Mass General Brigham has filed a provisional U.S. patent application for its next-generation facial health algorithms and is expanding institutional review board-approved public validation studies, with the health system also confirming it is opening the research to a public online study.
Independent Experts Urge Caution Before Clinical Use
Not everyone is ready to see the tool used in treatment decisions just yet. AI Weekly notes that outside experts, including Dr. Wilbur Lam of Emory University, have emphasized that multi-center prospective clinical trials across diverse patient populations are necessary before facial AI tools can be safely deployed in clinical care. Researchers envision the technology eventually helping oncologists quantify a patient's frailty before prescribing high-toxicity therapies, but AI Weekly's reporting underscores that real-world deployment hinges on trials that have not yet happened.
For now, the tool remains a research project rather than a bedside instrument. Mass General Brigham has filed a patent application and is opening public studies. Oncologists involved in the work frame it as a way to support personalized treatment decisions, not replace physician judgment.
