▲ AI model's prediction of cancer survival probability and biological age estimation
An artificial intelligence (AI) tool has been developed to predict the survival risk of cancer patients using just a single facial photograph.
Bloomberg reported on the 19th local time that the research findings detailing this tool were published in the medical journal The Lancet Digital Health.
Bloomberg explained that the tool has the potential to complement physicians' intuition and aid in treatment decisions, and could be particularly useful when medical information is scarce.
The AI developed by the researchers analyzes facial photos of cancer patients to generate a risk score, which is then used to predict the patient's survival risk.
The analysis showed that patients with the highest risk scores had a higher likelihood of dying within 90 days.
Additionally, the AI estimated patients' ages with a margin of error of about five years and reliably classified patients into high-risk and low-risk groups.
Actual survival risks also differed between the two groups.
The tool was confirmed to function across patient groups with diverse ages, ethnicities, genders, and types of cancer.
Dr. Raymond Mak, a radiation oncologist at the Mass General Brigham Cancer Center at Harvard Medical School and co-lead of the study, explained that while doctors are trained to assess health status based on a patient's appearance, they are generally not skilled at estimating the remaining life expectancy of terminally ill patients.
He stated that even a single facial photo can provide some insight into a patient's physiological reserve—the body's ability to withstand stress or disease—and their risk of mortality, adding that he was surprised by the predictive performance of the AI.
The researchers also found that the AI focused particularly on the bridge of the nose and the area directly beneath the eyes.
Dr. Mak noted that the ancient Greek physician Hippocrates described sunken eyes, pinched noses, and hollowed temples as early physical signs indicating imminent death.
The researchers believe the tool can be used to complement physicians' judgments and assist in treatment decisions.
They explained that it could be particularly useful in situations where medical information is lacking.
Dr. Mak emphasized that doctors tend to over-treat terminally ill patients, and the AI would serve as a means to provide doctors with an additional piece of information for their judgment rather than restricting treatment.
However, some point out that additional validation is needed before it can be applied in actual medical settings.
Miroslaw Bober, a professor of image processing and machine learning at the University of Surrey in the UK who did not participate in the study, praised the research as interesting and solid, highlighting that the results were validated on an independent cohort of about 5,000 Dutch cancer patients.
However, he pointed out that it has not been proven whether the AI captures information that doctors miss or provides practically useful information in addition to existing clinical data.
Professor Bober explained that the study shows a statistical link between the facial appearance and survival rates of cancer patients receiving radiation therapy, and that the AI can classify patients by risk level.
He emphasized, however, that predicting results and explaining their causes are two different things.
He added that the expression "biological age" used in the paper must be interpreted with caution.
This is because the AI does not directly measure the actual biological aging process, but rather estimates age based on facial appearance.
Meanwhile, cases utilizing AI in the detection, diagnosis, and treatment of cancer are steadily increasing.
The scope of application is expanding from diagnostic assistance through medical image reading to predicting cancer risk by analyzing genetic factors and lifestyle habits.
While this study also demonstrated the potential to predict cancer patients' survival risks using only facial photographs, how much additional help it can provide in actual clinical practice remains to be further verified.
(Photo: Publicly released photo from the Mass General Brigham website, Yonhap News)
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.