AI Predicts Cancer Death Risk From a Single Facial Photo
An AI has emerged that predicts the death risk of cancer patients from a single facial photo. The tool consists of two AI models — one estimates the patient's age, while the other calculates a risk score. The survival-prediction AI analyzes the facial photo and assigns a risk score from 0 to 1. The study was conducted by researchers at Mass General Brigham, a Harvard Medical School-affiliated medical research institution, and the results were published in the medical journal The Lancet Digital Health. Raymond Mak, a radiation oncologist who participated in the study, said, "Using AI, a single facial photo can provide a fair sense of a patient's physiological reserve and risk of death."
The analysis showed that the higher the risk score, the greater the risk of death. Half of the patients with a risk score of 0.9 or higher died within 90 days of radiotherapy. Survival gaps based on the score were also stark: the five-year survival rate was 78.1% for the lowest-scoring group, but only 17.7% for the highest. Patients whose faces made them appear at least five years older than their actual age had a 21% higher risk of death than those who did not. This predictive performance held true across groups differing in age, sex, race, and cancer type. A separate validation involving about 4,900 cancer patients in the Netherlands also confirmed the association between the facial analysis results and survival rates.
The researchers paid particular attention to which parts of the face the AI examined. The survival-prediction AI primarily based its risk assessment on the bridge of the nose and the area directly below the eyes, while the age-estimation AI mainly scrutinized nasolabial folds, the forehead, and the temples. However, the specific reasons why the AI focuses on these areas remain unknown. Professor Mak said that after seeing these results, he was reminded of the writings of Hippocrates, who described features such as sunken eyes and temples and a sharpened nose as physical signs that death was imminent.
Miroslaw Bober, a professor of image processing and machine learning at the University of Surrey in the UK, an outside expert who did not take part in the study, called it "an interesting and robust study." Bober highlighted as a key strength the fact that the predictive performance was validated using an independent group of about 4,900 cancer patients in the Netherlands.
That said, Bober pointed out that it remains unconfirmed whether the AI provides new information beyond the judgment of existing medical staff. "This study did not test whether the AI captures something that physicians cannot detect," he said, adding, "It has also not been proven that the AI provides additional clinically useful information." Regarding the "biological age" mentioned in the study, Bober noted that the AI did not directly measure the body's actual aging process but merely estimated age based on facial appearance. He emphasized that while the study demonstrates a statistical association between facial appearance and survival among cancer patients undergoing radiotherapy, the fact that the AI can stratify patients' death risk does not mean it can accurately predict an individual's survival time or explain its causes.
