Mass General Brigham Trains 40 Million-Image AI to Predict Cancer Survival From Faces
Updated
Updated · The Guardian · Oct 23
Mass General Brigham Trains 40 Million-Image AI to Predict Cancer Survival From Faces
3 articles · Updated · The Guardian · Oct 23
Summary
Mass General Brigham researchers built FAHR-Face on more than 40 million facial images, including nearly 35,000 photos of cancer patients, to estimate biological age and predict survival.
The Lancet Digital Health study found AI-estimated facial age was significantly linked to survival in 18 of 28 cancer categories, with the survival model focusing on areas below the eyes and around the nose.
The results remain preliminary because the models were trained on retrospective associations, still need testing against doctors' judgments across hospitals and patient groups, and did not show a significant adjusted survival link for Black patients.
Those limits add to broader concerns that medical AI can encode racial bias and may be vulnerable to selfie variables such as lighting, angle, expression and cosmetic changes, leaving its clinical usefulness unproven.