AI Tool Can Predict Cancer Survival From Selfie, Study Suggests

An artificial intelligence tool capable of predicting cancer patient survival rates by analyzing facial photographs has been developed by U.S. researchers. The AI, named FaceSurvival, identifies individuals at the greatest risk of dying, potentially informing treatment decisions.

In tests, the AI identified a group of patients at high risk of death, with more than half of them dying within 90 days. In contrast, only 2 percent of patients categorized in the lowest-risk group died during the same period. Experts suggest this technology could help clinicians determine which patients are suitable for aggressive cancer treatments and which might require a more conservative approach.

The development builds upon earlier research that indicated cancer patients who appear older than their chronological age have a higher likelihood of mortality. Instead of estimating biological age directly, this new AI focuses on subtle facial cues that correlate with survival outcomes.

"Clinicians already intuitively assess patient health based on appearance," said Dr. Grace Lee, a co-author of the study and an expert in colon and rectal surgery. "Our tool provides a way to standardize and quantify this intuitive practice, potentially enhancing prognostic accuracy and clinical decision making."

To train the AI, researchers utilized over 40 million publicly available facial photographs. FaceSurvival was then taught to predict mortality using images from more than 34,000 cancer patients. When evaluated on a separate group of 3,820 patients, the tool assigned each a risk score ranging from zero to one.

Among the 122 patients who received the highest scores, 64 died within three months. This contrasts with 20 deaths among the 975 patients in the lowest-risk group. Even after accounting for factors like age, sex, and cancer type, individuals in the highest-risk quarter were more than three times as likely to die compared to those in the lowest-risk quarter. The study's findings were further validated on nearly 4,900 cancer patients in the Netherlands.

A secondary tool, FAHR-FaceAge, was also developed to estimate a person's biological age from their face. Biological age reflects the body's aging process, distinct from chronological age. In previous work, the team demonstrated this concept using photographs of actors Paul Rudd and Wilford Brimley, both taken at age 50. Rudd, now 57, was estimated to have a biological age of 42.6, while Brimley, who died at 85, was estimated to have a biological age of 69.

The latest study applied the biological-age tool to over 38,000 cancer patients. Those whose faces appeared at least five years older than their actual age had a 21 percent higher risk of death, while those who looked at least five years younger had a 17 percent lower risk. Factors like smoking, heavy drinking, and drug use were also associated with an older-looking face, with current smokers showing the strongest link to an increased biological age.

While the researchers believe these tools could eventually offer a faster and more cost-effective way for doctors to assess patients, the technology is still considered experimental. Further studies are necessary before its integration into clinical practice. Scientists are also investigating the impact of cosmetic surgery on the AI's accuracy, and whether using these tools to guide treatment will ultimately improve survival outcomes remains an open question.