AI predicts cancer treatment survival from a single face photo

Researchers at Mass General Brigham built FAHR-Face, an AI model that estimates biological age and survival odds for cancer patients from one facial photograph. Age estimates had a mean error of 5.1 years, and 52% of patients in the highest-risk group died within 90 days versus 2.1% in the lowest-risk group.

• Model trained on over 42 million face photos using masked autoencoders
• Biological age estimated with a mean error of 5.1 years
• 52% died within 90 days in top-risk group vs 2.1% in lowest
• Study published in The Lancet Digital Health