Researchers at Mass General Brigham have built an AI model that can estimate a cancer patient’s chances of survival from a single photograph of their face.
The study, published in The Lancet Digital Health on October 19, is the first to show an AI system predicting survival directly from a face photo rather than from scans or clinical tests.
Two tools built on one foundation model
The model is called FAHR-Face, short for Foundation Artificial Intelligence for Health Recognition. It was trained on more than 40 million images of faces across a range of ages and ethnicities, and the team then adapted it into two separate tools.
FAHR-FaceAge estimates biological age, meaning how old a person looks rather than their actual age. FAHR-FaceSurvival skips the age step and predicts survival outcomes directly from the photograph. The two tools capture different information, and combining them improved the accuracy of survival predictions in testing.
What the results showed
In a test group of 3,820 cancer patients, more than half of the patients given the highest risk scores died within 90 days of their photograph being taken. In a much larger group of more than 38,000 patients, those who looked older than their real age faced a higher risk of death.
Among patients receiving radiotherapy, appearing at least five years older than their chronological age was linked to a 21 percent higher risk of death over an average follow-up of four years.
Appearing five or more years younger came with a 17 percent lower risk. The age estimates themselves were accurate to within about five years, and the model held up across different ages, races, genders and types of cancer.
“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 Mass General Brigham Cancer Institute and the study’s corresponding author.
Where the AI looks
Using a technique called attention mapping, the researchers tracked which parts of the face each tool relied on. For age, the model focused on the creases running from the sides of the nose to the corners of the mouth, the forehead just above the eye sockets and the temples.
For survival risk, it focused on the region below the eyes and around the bridge of the nose. Why those areas carry survival signals is still unclear. The age tool also stayed reliable across changes in pose, lighting, facial expression and makeup, and it rated people who smoked, drank heavily or used drugs as looking biologically older.
Built to support doctors, not replace them
“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,” Mak said. The team frames the tools as a way to make that assessment more quantitative, consistent and scalable, supporting personalized treatment decisions rather than replacing physician judgment.
Doctors already judge how a patient looks when they walk into a room, but that assessment is subjective and doctors are generally not very good at estimating end-of-life expectancy. A photograph-based score could give that instinct a measurable form, especially in settings where detailed medical data is scarce.
Why caution is still needed
The study was a model development and validation exercise, not a clinical trial. The researchers did not test whether doctors using the scores actually made better treatment decisions or whether the tools changed outcomes for patients.
Miroslaw Bober, a professor of video processing and machine learning at the University of Surrey, said the tool is not ready for clinical use. Mass General Brigham has filed a provisional US patent application on next-generation facial health algorithms and is expanding the research through a public online study approved by an institutional review board.
The study builds on the team’s earlier FaceAge system, published in 2025, which estimated biological age from face photographs in more than 6,000 cancer patients and outperformed clinicians at predicting short-term life expectancy for patients receiving palliative radiotherapy. It also lands amid a wider surge of AI in medicine, visible in the healthcare AI startups building tools for imaging, diagnostics and clinical care.
