AI tool predicts cancer survival chances with single face photo
Trained on more than 40 million face images, the AI model gives cancer patients risk scores.
- The AI tool could estimate age with an error of about 5 years and reliably place patients into groups at higher or lower risk of death. PHOTO: UNSPLASH
Researchers have developed an artificial intelligence (AI) tool that can estimate how long cancer patients might live by looking at a single photo of their face. The tool could add to doctors' own judgments and help them make treatment decisions, especially when little medical information is available.
The new model was trained to estimate a person's age and, for people with cancer, their chances of survival.
The AI was trained using more than 40 million face images and photos of over 30,000 cancer patients. Many of these patients had already died.
The tool gave each patient a risk score. According to the study published in The Lancet Digital Health, patients with the highest scores were more likely to die within 90 days.
The researchers said this was the first study of its kind. Raymond Mak, a cancer doctor who specialises in radiation treatment at Mass General Brigham Cancer Institute and one of the study's authors, said doctors are trained to judge patients by how they look. However, he said they are generally not very good at estimating how much time a patient has left to live.
“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 could estimate a person's age with an error of about 5 years. It could also reliably sort patients into groups at higher or lower risk of death, and the groups had different chances of dying. The model worked across different ages, races, genders and cancer types.
AI tools are being tested more and more for cancer detection and treatment. Some help doctors find diseases by examining medical scans. Others estimate a person's risk of developing cancer by looking at genetic and lifestyle risk factors.
Despite these promising results, Miroslaw Bober, a professor of video processing and machine learning at the University of Surrey, said the tool is not ready to be used in medical care.
“Predicting an outcome is not the same as explaining it,” said Bober, who did not take part 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 researchers need to compare the tool with doctors' own judgments and test it in different hospitals and groups of patients.
Mak, who developed the tool with his team, said he was amazed by how well it predicted survival.
The researchers found that the algorithm mainly looked at the bridge of the nose and the area just below the eyes when making its predictions.
Mak looked into the history of this idea and found that Hippocrates was the first to describe sunken eyes and temples, along with a sharp nose, as early physical signs that death might be near.
Mak said the tool is unlikely to be used to refuse patients medicine or limit their access to treatment. “We know that doctors at the end of life tend to overtreat,” he said. He sees the tool as “a way for doctors to have one more piece of information”. BLOOMBERG
Decoding Asia newsletter: your guide to understanding Asia in a changing world. Sign up to receive the Decoding Asia newsletter in your inbox for free.