Facial AI scores reveal complementary cancer survival signals in new study
Combining age and survival-risk estimates improved survival-model fit. But the survival tool learned from a majority-white patient population, and clinical testing remains ahead.
Listen to this story
The audio brief
Story brief
3 key pointsPublished October 19 in The Lancet Digital Health, the study suggests facial-image models could add distinct survival information, but it does not establish a clinical tool. The results come with important limits: the training population was majority white and drawn from the Boston area, and researchers say broader data and clinical testing are needed. Because an inaccurate risk estimate could affect treatment access, the proposed role is to support—not replace—physician judgment, with possible misuse by insurers or employers also flagged.
- 01
FAHR-Face was trained on more than 40 million facial images and adapted to estimate biological age and cancer survival risk.
- 02
In a test group of 3,820 patients, the highest-risk subgroup had substantial mortality within 90 days of being photographed; this is not an overall-accuracy measure.
- 03
In a separate group of more than 38,000 patients, estimates placing patients older than their chronological age were associated with higher mortality risk.
Two AI tools analyzing facial photographs captured different signals linked to cancer patients’ survival, Mass General Brigham researchers report. Combining their biological-age and survival-risk estimates improved survival-model fit in an external patient group. The findings, published October 19 in The Lancet Digital Health, remain research—not a tool ready for clinical use.
The two tools’ predictions were only weakly correlated, and each remained independently associated with survival when analyzed together. That suggests they captured complementary health information rather than simply duplicating one another.
Scientific American reported a striking result from a test group of 3,820 cancer patients. The highest-risk subgroup had substantial mortality within 90 days of being photographed. That subgroup result is not a measure of the model’s overall accuracy.
In a separate test group of more than 38,000 cancer patients, those estimated to look older than their chronological age also had a higher risk of dying.
Treatment decisions raise the stakes
Mak’s goal is to make assessments of apparent health and frailty more consistent, supporting treatment decisions rather than replacing physician judgment. New York University bioethicist Matthew Liao, who was not involved in the study, warned that a mistaken assessment could lead to needed care being withheld. He also raised potential misuse by insurers or employers.
One model, two facial measures
Researchers trained FAHR-Face on more than 40 million facial images. This foundation model learns a general representation of faces that can be adapted to different health questions, rather than requiring each tool to be trained from scratch.
- FAHR-FaceAge estimates biological age. The gap between that estimate and a patient’s age in years was independently associated with survival.
- FAHR-FaceSurvival predicts survival risk directly from a cancer patient’s photograph, rather than using an age estimate as the health signal.
Attention mapping, which highlights influential image regions, showed different patterns. Age estimates drew on creases between the nose and mouth and parts of the forehead. Survival estimates focused below the eyes and around the nasal bridge. Why those regions correlate with mortality remains unclear.
Sources
- massgeneralbrigham.orgAI facial health recognition tools help physicians assess age and survival risk to guide personalized treatment | Mass General Brigham
- scientificamerican.comCan AI predict whether you’ll die from cancer from just an image of your face?
Reader comments
Newest comments first. Replies stay oldest first.