AI Model Trained on 40 Million Faces Predicts Cancer Mortality Risk From Photos

BOSTON — Researchers at Mass General Brigham have developed an artificial intelligence system that analyzes a single facial photograph to estimate biological age and predict survival outcomes in patients with cancer, according to a study published this week in The Lancet Digital Health.

The foundation model, called FAHR-Face, was trained on more than 40 million facial images. From it, the team created two tools: FAHR-FaceAge, which estimates a person's biological age, and FAHR-FaceSurvival, which predicts mortality risk directly from a photograph. The researchers said this is the first demonstration that an AI model can directly predict cancer survival from a simple face photograph.

"FAHR-FaceSurvival advances the field of facial health recognition AI a step further by learning survival risk directly from facial photographs of patients with cancer," said Hugo Aerts, director of the Artificial Intelligence in Medicine program at Mass General Brigham and a co-senior author of the study. "To our knowledge, this is the first demonstration that an AI model can directly predict cancer survival from a simple photograph."

The models build on earlier work by the same group on a tool called FaceAge, which estimated biological age from photographs. Biological age reflects physiological condition rather than chronological age. In the new study, patients with cancer whose estimated biological age was at least five years older than their chronological age had a 21 percent higher risk of death during follow-up than other patients, after adjustment for age, sex, cancer type and treatment factors. Those who appeared five or more years younger had a 17 percent lower risk.

In one analysis involving 3,820 cancer patients, more than half of those assigned the highest risk scores by FAHR-FaceSurvival — 64 of 122 — died within 90 days of the photograph. Among 975 patients in the lowest-risk group, about 2 percent died within the same period. In a larger cohort of more than 38,000 patients treated with radiotherapy, the association between an older-looking face and higher mortality held after statistical adjustments.

Raymond Mak, a radiation oncologist at Mass General Brigham Cancer Institute and a co-senior author, said the tools capture information that clinicians have long assessed subjectively. "We're trained in medical school to walk into a room, look at the patient and make an assessment — an impression of how healthy, frail or sick they are," Mak said. He added that the mortality-risk model focuses on different facial regions than the age-estimation model, including the area below the eyes and around the nasal bridge, while FaceAge emphasizes nasolabial folds, the forehead above the eyes and the temples.

The two measures were only weakly correlated with each other, and each remained independently associated with survival. Combining them improved the performance of prognostic models in an external validation group. The researchers also reported that the tools performed consistently across common variations in photo quality and lighting.

Mak said the goal is to make health assessment from a photograph "more quantitative, consistent and scalable, so it can support personalized decisions, not replace medical judgment." He noted potential uses in evaluating older or frail patients before intensive treatments such as radiation.

An independent expert, Pratik Shah, a professor of pathology at the University of California, Irvine, who directs its Machine Learning for Health Research Program and was not involved in the study, called the work an innovative use of AI. He said the large training set and the size of the cancer-patient cohort distinguish it from earlier facial-age tools. Shah cautioned that the findings come from historical data rather than a prospective clinical trial and that performance varied by cancer type. "The facial-age model showed significant survival associations in 18 of 28 cancer categories," he said. "The model may not perform consistently across cancer types."

The Mass General Brigham team has launched clinical trials to test the tools further. One trial at Brigham and Women's Hospital is enrolling patients ages 65 and older with early-stage lung cancer to examine whether the technology can help predict outcomes before radiation therapy. A research study is also open for public participation at faceage.bwh.harvard.edu, where individuals can upload photographs.

Earlier research by the group, published in Nature Communications, examined changes in facial appearance over time among 2,279 cancer patients who had photographs taken at different points during treatment. Patients whose faces appeared to age more rapidly had lower survival probabilities. The median rate of facial aging outpaced chronological aging by about 40 percent in that cohort.

The new models were trained without an initial specific task on the large set of facial images, then fine-tuned for biological-age estimation and survival prediction. The age-estimation component was refined on hundreds of thousands of portraits with known chronological ages and showed an average error of roughly five years on held-out test images.

Researchers said lifestyle factors such as heavy alcohol use, drug use and smoking were associated with higher estimated biological ages in related analyses. The tools are not intended as standalone diagnostic devices. The study authors emphasized that any clinical application would require further prospective validation and integration with existing clinical data.

The research was conducted at Mass General Brigham in Boston, an integrated academic healthcare system. The findings appear in the October 2026 issue of The Lancet Digital Health. No commercial availability of the tools has been announced.