A simple facial photograph may reveal clues about biological age and survival among people with cancer, according to research from Mass General Brigham in Boston. Researchers developed artificial intelligence models that analyze facial features to estimate health related risks.

The research, published Monday in The Lancet Digital Health, suggests facial analysis could eventually provide doctors with another tool when evaluating survival prospects and considering treatment approaches. The technology is not intended to replace physicians or existing clinical assessments.

“I was pretty surprised to find out that we could pick up signs of mortality risk from just a facial photograph,” said study author Dr. Raymond Mak, a radiation oncologist at Mass General Brigham Cancer Institute and a faculty member in the Artificial Intelligence in Medicine program at Mass General Brigham.

The artificial intelligence models were trained on more than 40 million facial images. They assess different parts of a person’s face to estimate biological age, which reflects how the body has aged at the cellular and physical level rather than simply counting years lived.

For age estimation, the algorithm focuses on areas around the temples and forehead, above the eyes, and along the nasolabial fold connecting the nose to the corner of the mouth. The mortality prediction algorithm examines different areas, including the bridge of the nose and the region beneath the eyes.

Those facial regions recalled a description attributed to Greek physician and philosopher Hippocrates in the Book of Prognostics, written around 400 B.C. He described “the nose sharp, the eyes sunken, the temples fallen in” as signs associated with muscle wasting and mortality risk.

“To think that our algorithms might be replicating that wisdom of Hippocrates from millennia ago was mind-blowing to us,” Mak said. He stressed that the algorithms had no awareness of the ancient text because they were trained using facial photographs paired with age or mortality information.

Mak and his colleagues first created the Foundation Artificial Intelligence for Health Recognition-Face, known as FAHR-Face. The model used more than 40 million facial images representing diverse ages and ethnic backgrounds before researchers adapted it for more specific tasks.

The FAHR-FaceAge model estimates biological age from one photograph. A second model, FAHR-FaceSurvival, predicts a cancer patient’s survival prospects by combining facial features with other clinical information.

Among 122 cancer patients identified by FAHR-FaceSurvival as having a high risk of death, 52.5% died within 90 days of the assessment. By comparison, about 2% of the 975 patients placed in the model’s lowest risk group died during that period.

Researchers also evaluated results from FAHR-FaceAge. Among 1,060 patients whose estimated biological age exceeded their chronological age by more than 20 years, 21% died within 90 days.

The corresponding figure was about 6% among 4,195 patients whose estimated biological age was at least 10 years younger than their chronological age. Researchers reported that facial changes involving surgery, Botox, other injectables or makeup did not affect the models’ performance.

Mak said accelerated aging or higher mortality scores might eventually help identify patients who could benefit from additional testing for serious conditions. Although the study included only people with cancer, he said further research could examine whether similar findings appear among otherwise healthy people interested in longevity.

Outside experts described the findings as compelling but cautioned that clinical usefulness remains unproven. Dr. Wilbur Lam, research chair and professor of pediatrics and biomedical engineering at Emory University, said the technology requires prospective evaluation and testing among specific populations before clinical implementation.

Lam, who was not involved in the research, said questions remain about performance among people with particular facial features, skin colors and age groups. Those effects would need prospective testing before the technology could be introduced into clinical practice.

Miroslaw Bober, professor of video processing and machine intelligence at the University of Surrey, said the study offered convincing evidence of a statistical association between facial appearance and survival among cancer patients receiving radiotherapy. However, he cautioned that predicting an outcome does not explain it.

Bober noted that the model estimates age from facial appearance instead of directly measuring an underlying biological aging process. He also said the research did not establish whether the system detects information that physicians would otherwise miss.

The study did not test whether the predictions add clinically useful information beyond doctors’ assessments of general health and fitness, according to Bober. He described the technology as a promising proof of concept that still requires testing against physicians’ judgments and across different hospitals and patient groups.