AI tool predicts cancer survival and biological age from a single face photo

An artificial intelligence system can now estimate a person's biological age and predict the survival chances of cancer patients using nothing more than a single photograph of their face.

Developed by researchers at Mass General Brigham in the United States, the foundation model, named FAHR-Face, was built by training algorithms on more than 40 million facial images alongside medical photographs of over 30,000 cancer patients. The findings, published in The Lancet Digital Health, show that the software can sort patients into distinct mortality risk categories, with those assigned the highest scores facing a sharply higher risk of dying within 90 days.

Clinicians often rely on visual assessment to gauge how frail a patient is before recommending aggressive therapies such as chemotherapy, extensive surgery, or radiotherapy. However, doctors are rarely precise when calculating end-of-life trajectories.

"As the population ages, physicians need better ways to estimate physical reserve and frailty before recommending intensive treatments," said Dr Raymond Mak, a radiation oncologist at the Mass General Brigham Cancer Institute and corresponding author of the study. He noted that the objective was to make visual evaluation "more quantitative, consistent and scalable so it can support personalized treatment decisions rather than replace physician judgment."

How the two tools evaluate the face

The research team derived two distinct algorithms from the core foundation architecture: FAHR-FaceAge, which calculates biological age, and FAHR-FaceSurvival, which estimates mortality risk.

Although both tools emerged from the same base model, they look at entirely different parts of the human face and provide weakly correlated, complementary data.

Dr Mak noted that the survival model's focus on sunken features around the bridge of the nose and the lower eye sockets closely mirrors clinical observations made millennia ago. Hippocrates famously identified sharp facial contours and hollowed eyes as clear physical precursors to death.

The researchers also observed that external lifestyle factors significantly accelerated apparent biological age. Regular tobacco use produced the most pronounced effect, with active smokers appearing noticeably older than former smokers, who in turn showed more biological aging than lifelong non-smokers. Heavy alcohol intake and drug use also drove up biological age estimates.

Independent experts urge clinical validation

Independent scientists have warned that significant work remains before such tools can enter routine hospital care. Professor Miroslaw Bober, who specialises in machine learning and video processing at the University of Surrey and was not involved in the work, stressed that predicting an outcome does not explain its underlying clinical cause.

Bober pointed out that the trial did not compare the algorithm directly against traditional clinical evaluations of physical fitness and general health. Before oncologists can rely on automated visual scoring, the software must undergo validation across different health systems, demographic populations, and hospital workflows.

The Mass General Brigham team emphasised that the software is intended to curb end-of-life overtreatment rather than restrict access to essential therapies. To broaden validation across diverse global populations, the researchers have launched the public FaceAge Research Study online, allowing participants to submit photographs and assist in refining the underlying demographic data.