This AI model is trained to predict cancer survival from a single photo

Researchers are studying how artificial intelligence (AI) can use a single photo of a person's face to predict how likely they are to survive cancer.

Researchers at Mass General Brigham in Boston created a large AI model called FAHR-Face, which stands for Foundation Artificial Intelligence for Health Recognition. They trained it using more than 40 million facial images.

An earlier version of the model, called FAHR-FaceAge, predicted a person's biological age. This means how much a person's body has aged at the cellular level, rather than how old they are in years. The new findings, published Monday in Lancet Digital Health, describe a more accurate version of this model.

The researchers also trained a second version of the model, called FAHR FaceSurvival, using 34,000 facial images of people with cancer. This specially trained model could predict a person's chances of surviving cancer using essentially just one selfie.

"FAHR-FaceSurvival advances the field of facial health recognition AI a step further by learning survival risk directly from facial photographs of patients with cancer," Hugo Aerts, one of the study's two senior authors, said in a news release. "To our knowledge, this is the first demonstration that an AI model can directly predict cancer survival from a simple photograph."

Aerts is also the director of Artificial Intelligence in Medicine at Mass General Brigham.

FaceSurvival predicted cancer survival by looking at facial features, such as lines running from the nose to the mouth and wrinkles around the eyes. About half of the cancer patients whose FaceSurvival scores were in the highest 10% died within three months. Among patients whose scores were in the lowest 10%, about 2% died within 90 days.

The researchers also found that people whose biological age was higher than their actual age were more likely to have worse cancer outcomes.

"As the population ages, physicians need better ways to estimate physical reserve and frailty before recommending intensive treatments such as radiation, surgery or systemic therapy," Dr. Raymond Mak, a radiation oncologist at Mass General Brigham Cancer Institute, said in the news release. "Our goal is to make face photograph-based health assessment more quantitative, consistent and scalable so it can support personalized treatment decisions rather than replace physician judgment."

The researchers said lifestyle factors, such as smoking and drug use, affected people's facial features and made them look older in photos. However, scientists do not fully understand how AI uses facial images to estimate biological age and predict cancer outcomes.

"One of the fascinating unanswered questions is what exactly the algorithm is seeing," Suchi Saria, director of the John Hopkins University Machine Learning and Healthcare Lab, who was not involved in the study, told the Washington Post. "But we don't yet know what physiological features in those regions are carrying the signal. Understanding that would help us determine whether the model is discovering a genuinely new biomarker or quantifying visible manifestations of illness that clinicians already recognize."

Emma Pierson, an assistant professor of computer science and a core faculty member at the University of California, Berkeley, told the Washington Post that she was concerned that the model might be biased against certain racial groups.

"Any time you train something on someone's face, you are implicitly including patient race in the model's predictions," said Pierson, who was not involved in the research.

Because different racial groups have historically had unequal access to quality health care, their health outcomes can differ. Training AI using facial images could mean "encoding racial bias into your model," Pierson said.