AI tool predicts cancer survival chances with single face photo, study finds

• Researchers say an AI tool can predict cancer patients’ survival chances from a single face photo, using risk scores to separate higher- and lower-risk groups.
• The model was trained on more than 40 million face images and over 30,000 cancer patient photos, and it estimated age with about five years’ error.
• Experts said it is not ready for clinical use and must be tested against doctors’ judgement, but the team says it may give doctors “one more piece of information”.

NEW YORK - Researchers built an AI tool that can predict cancer patients’ survival chances by using a single photograph of their face, complementing doctors’ gut instincts to potentially guide treatment decisions, especially when medical data is scarce. 

The new model was trained to estimate a person’s age or, for those with cancer, their survival chances.

The artificial intelligence was trained on more than 40 million images of people’s faces as well as photographs of more than 30,000 cancer patients, many of whom had already died.

The tool gave patients risk scores – those with the highest scores were more likely to die within 90 days, according to the research published in The Lancet Digital Health on Oct 19.

The scientists say the findings are a first.

Doctors are trained to assess patients just by how they look, said Raymond Mak, a radiation oncologist at Mass General Brigham Cancer Institute and one of the study’s authors, but they are “in general not very good at estimating end-of-life expectancy”.

“We think that using AI, we can capture some of that physiological reserve, risk for mortality, just from a face photo,” Mak said in an interview. 

The AI tool showed it could estimate age with a margin of error of about five years and reliably allocate patients to higher- or lower-risk groups, which subsequently had different risks of death.

The model worked across different ages, races, genders and types of cancer. 

AI tools are increasingly being tested for cancer detection and treatment, from helping to diagnose patients by reading scans to predicting the risk of a patient developing cancer by analysing genetic and lifestyle risk factors.

Despite the promising results, Miroslaw Bober, professor of video processing and machine learning at the University of Surrey, said the tool is not ready for clinical use. 

“Predicting an outcome is not the same as explaining it,” said Bober, who was not involved in the research.

“It has not shown that these predictions add clinically useful information beyond doctors’ assessments of a patient’s general health and fitness, which were not measured in the study,” he said.

Bober added that the tool needs to be tested against doctors’ own judgment as well as across different hospitals and patient groups.

Mak, who developed the tool with his team, said he was stunned by how well it worked for survival.

The algorithm focused mostly on the bridge of the nose and the area just beneath the eyes to make its assessment, they found.

He did some digging and found that Hippocrates was first to describe how sunken eyes and temples as well as a sharp nose were among the early physical signs of impending death. 

The tool is unlikely to be used to deny or limit patients’ access to medicine, according to Mak.

“We know that doctors at the end of life tend to overtreat,” he said. He views it as “a way for doctors to have one more piece of information”. Bloomberg