Doctors have developed a cancer survival AI tool that can predict a cancer patient’s chances of survival from a single photograph of their face, in one of the most comprehensive demonstrations yet of how AI analysis of a selfie could guide medical care. The system was custom-built by experts at Mass General Brigham, the nonprofit academic healthcare system in Massachusetts, and the details were published in The Lancet Digital Health on October 19.
The AI model was trained on more than 40 million images of human faces as well as photographs of more than 30,000 cancer patients, many of whom had already died. From those inputs, the system assigns patients risk scores — those with the highest scores were more likely to die within 90 days. The scientists say the findings are a first of their kind.
“We think that using AI, we can capture some of that physiological reserve, risk for mortality, just from a face photo,” said Raymond Mak, a radiation oncologist at the Mass General Brigham Cancer Institute and one of the study’s authors. Mak noted that doctors are trained to assess patients just by how they look but are “in general not very good at estimating end-of-life expectancy”.
How the Cancer Survival AI Was Built
The researchers first trained a large foundation model — named FAHR, for Foundation Artificial Intelligence for Health Recognition — on tens of millions of face images without giving it a specific task, then adapted it to two health-related outcomes. They used the shared model to derive two complementary tools: FAHR-FaceAge, which estimates a patient’s biological age, and FAHR-FaceSurvival, which predicts survival from a facial photograph directly rather than inferring it through biological age.
Although the two algorithms were derived from the same model, they captured different and complementary information about a patient’s health. The study reported that the difference between estimated biological age and chronological age was independently associated with survival, and that combining the two tools improved the model’s performance in an external cohort of cancer patients.
The AI tool showed it could estimate age with a margin of error of about five years and reliably allocate patients to higher- or lower-risk groups that subsequently had different risks of death. The model worked consistently across different ages, races, genders and types of cancer, the researchers reported.
What the Cancer Survival AI Revealed
In a test group of 3,820 cancer patients, more than half of those with the highest risk score — 64 out of 122 — died within just 90 days of when their photograph was taken. In a separate test group that included more than 38,000 cancer patients, those who looked older than their chronological age also had a higher risk of dying.
Using a technique the team calls “attention mapping”, the researchers tracked where the algorithm was looking. For age prediction, the AI tended to base its predictions on facial areas such as the nasolabial folds — the creases between the nose and mouth — and parts of the forehead. For risk of death, it focused on the bridge of the nose and areas below the eyes, though it remains unclear why those areas specifically correlated with mortality risk.
The researchers also found that lifestyle influenced facial ageing: heavy alcohol use, drug use and smoking were associated with looking biologically older, with smoking showing the strongest pattern. Smokers looked older than former smokers, who in turn looked older than people who never smoked. The findings were consistent across photo variations including pose, lighting, facial expression and makeup, the team said.
Limits of the Cancer Survival AI
Despite the promising results, experts cautioned that the tool is not ready for clinical use. Miroslaw Bober, professor of video processing and machine learning at the University of Surrey, said the findings need further validation before doctors could rely on them.
The study was computational rather than a clinical trial: it developed and validated prediction models but did not test whether physicians using the scores made better treatment decisions or whether the tools changed patient outcomes. “Our goal is to make face photograph-based health assessment more quantitative, consistent and scalable so it can support personalised treatment decisions rather than replace physician judgment,” Mak said.
Hugo Aerts, professor at Harvard University and director of the AI in medicine programme at Mass General Brigham, said the work “advances the field of facial health recognition AI a step further by learning survival risk directly from facial photographs of patients with cancer”. To his knowledge, he said, “this is the first demonstration that an AI model can directly predict cancer survival from a simple photograph” — a claim the study authors echo in their Lancet Digital Health paper.