AI Reads Facial Images to Forecast Cancer Outcomes and Aging

WHO:

Newser - Researchers at Mass General Brigham developed an AI model called FAHR-FaceSurvival.

Newser - Outside experts described the work as “intriguing,” according to The Washington Post, and said it needs to show that it adds information beyond what doctors can already see and avoids encoding racial or other biases.

NBC News Health - Researchers at Mass General Brigham developed AI models for estimating cancer survival and biological age.

NBC News Health - Jared Perlo, NBC News artificial intelligence reporter, and Dr. Akshay Syal, NBC News medical contributor, explained how the models work.

WHAT:

Newser - FAHR-FaceSurvival estimates cancer patients’ mortality risk using a facial photograph.

Newser - Among patients in the model’s top 10% risk group, about half died within 90 days; among those in its lowest 10% risk group, roughly 2% died within 90 days.

NBC News Health - The AI models may predict cancer survival and estimate biological age, but researchers said they are not ready for clinical use.

WHERE:

Newser - The research was conducted by researchers at Mass General Brigham.

NBC News Health - The models were developed by researchers at Mass General Brigham.

WHY:

Newser - Researchers developed FAHR-FaceSurvival to estimate mortality risk from a cancer patient’s facial photograph.

HOW:

Newser - FAHR-FaceSurvival grew out of FaceAge, an earlier tool that estimates a person’s “biological age” and collects other information about a patient’s health.

Newser - Attention maps indicate that FAHR-FaceSurvival focuses on areas below the eyes and along the nose, but scientists do not yet know precisely which signals it reads.

NBC News Health - The models were trained on 40 million facial images.

(UPDATED October 23, 2026 08:12 UTC)

Researchers at Mass General Brigham have developed an artificial intelligence model that estimates cancer patients’ risk of dying from a facial photograph. Called FAHR-FaceSurvival, it grew out of FaceAge, an earlier tool designed to estimate biological age and gather other information about a patient’s health. The models were trained on 40 million facial images. In the model’s highest-risk group, about half of patients died within 90 days; in its lowest-risk group, roughly 2% died within that period. Attention maps suggest the system focuses on areas below the eyes and along the nose, though researchers do not yet know exactly what signals it detects. Outside experts called the work intriguing but said it must show that it adds information beyond what doctors can already observe and does not encode racial or other biases. Researchers say these models are not ready for clinical use.