Crow’s feet, crepey eyes and worry lines give us clues about a person’s age, but appearances can be deceiving when self-care, sun protection and genetics come into play. Other faces may show premature signs of aging due to stress, lifestyle choices or illness.
Researchers at Mass General Brigham in Boston are looking beyond aesthetics. They have developed an AI system trained on more than 40 million facial images that can estimate biological age and help predict survival among people with cancer.
“To our knowledge, this is the first demonstration that an AI model can directly predict cancer survival from a simple photograph,” said Hugo Aerts, director of the Artificial Intelligence in Medicine program at Mass General Brigham.
The technology could give doctors another way to assess a patient’s health before recommending demanding treatments such as chemotherapy, radiation or surgery. The research, published Oct. 19 in The Lancet Digital Health, adds to a growing field in which AI identifies subtle patterns that may reveal hidden health risks.
The Mass General Brigham system, called FAHR-Face, has two components. FAHR-FaceAge estimates a person’s biological age from facial appearance, which may differ from their chronological age. FAHR-FaceSurvival analyzes facial patterns to estimate the risk of death among patients with cancer.
Among patients treated with radiotherapy, those who appeared five or more years older than their actual age had a 21% higher risk of death during an average four-year follow-up, while those who appeared five or more years younger had a 17% lower risk.
“As the population ages, physicians need better ways to estimate physical reserve and frailty before recommending intensive treatments such as radiation, surgery or systemic therapy,” said Dr. Raymond Mak, a radiation oncologist at Mass General Brigham Cancer Institute and one of the study’s authors.
To build the system, researchers first trained an AI model on more than 40 million facial photographs, then used images from more than 34,000 patients with cancer to identify patterns associated with survival. They subsequently tested the system on separate patient data to evaluate its performance.
Rather than diagnosing cancer or predicting exactly how long someone will live, the system generates a risk score based on patterns learned from thousands of patients. Those with the highest FAHR-FaceSurvival scores had more than three times the risk of death as those with the lowest scores, even after researchers accounted for other factors.
The two components focus on different facial features. FAHR-FaceAge examines visible signs of aging, including creases between the nose and mouth and features around the forehead and temples, while FAHR-FaceSurvival concentrates on areas beneath the eyes and around the bridge of the nose. Combining the scores produced more accurate survival estimates than either score alone.
FAHR-FaceAge produced relatively consistent results despite variations in lighting, facial expressions, makeup and camera angles. Researchers also found associations between an older biological appearance and smoking, heavy alcohol consumption and drug use, with current smokers showing the strongest signs of accelerated facial aging.
The findings build on earlier research by the same team linking an older biological appearance to a higher risk of death among patients with cancer. Biological age can reflect a person’s overall physical condition, including the resilience needed to withstand intensive medical treatments.
The research is part of a broader effort to use AI to identify health risks before symptoms emerge. Some systems analyze medical histories and genetic information to predict disease, while others examine electrocardiograms, or ECGs, to identify people at greater risk of heart problems.
Researchers are also investigating whether mammograms can reveal cardiovascular risks in addition to detecting breast cancer, suggesting that routine medical tests may contain valuable information that conventional evaluations overlook.
Facial photographs could eventually become another source of medical insight, although the technology requires further testing. Because facial characteristics vary across populations, researchers must establish whether the system performs reliably across different ages, races and backgrounds. They also caution that the findings demonstrate a statistical association between facial appearance and survival, not an explanation of why that relationship exists.
For now, the Mass General Brigham researchers emphasize that the technology is not a substitute for clinical judgment. A photograph cannot determine a patient’s future, and an AI-generated risk score is not a diagnosis.
