A photo of the face may reveal the chances of surviving cancer treatment

A simple selfie may hold surprising clues about overall health and life expectancy, especially for cancer patients. Researchers at the Mass General Brigham Foundation in Boston, USA, have developed artificial intelligence models that can predict a cancer patient's biological age and chances of survival, based on a single image of his or her face.

The study, published in The Lancet Digital Health on Monday, suggests that facial analysis may constitute an additional tool that helps doctors evaluate patients' chances of survival and choose the most appropriate treatments for their condition. “I was quite surprised that we could detect indicators of death risk from just a facial image,” said the study's lead researcher, Dr. Raymond Mack, a radiation oncologist at MassGeneral's Brigham Cancer Institute and a faculty member in the institution's AI in Medicine program.

The AI ​​models have been trained on more than 40 million facial images and analyze different facial features to estimate a cancer patient's general health status and biological age. Unlike chronological age, which is measured by the number of years a person has lived, biological age reflects how much the body ages on a cellular and physical level.

Mack said that the age estimation algorithm analyzes the areas of the temples, forehead, and area above the eyes, in addition to what is known as the nasolabial fold, which is the line extending from the nose to the corner of the mouth, to estimate biological age. He added that, on the other hand, the algorithm for predicting the risk of death depends on completely different areas, including the bridge of the nose and the area under the eyes.

It is noteworthy that the same areas that artificial intelligence analyzes to estimate biological age and chances of survival were referred to by the Greek physician and philosopher Hippocrates in his book “The Book of Speculation,” which dates back to about 400 years BC. He described “a sharp nose, sunken eyes, and sunken temples” as signs of muscle wasting and a high risk of death.

“It was amazing to think that our algorithms could reproduce thousands of years of Hippocratic wisdom,” Mack said. He explained that the algorithms were not familiar with the ancient texts of Hippocrates, as they were trained only on images of faces and data on people’s deaths or ages.

He added that it "does not take into account texts, knowledge or wisdom accumulated on the Internet. So, it was really strange to hear the ideas of Hippocrates echoing across thousands of years in our algorithms."

Mack and his colleagues first developed an advanced AI model they called FAHR-Face, which stands for Foundation Artificial Intelligence for Health Recognition-Face, and trained on more than 40 million images of the faces of people of diverse ages and ethnic backgrounds. The researchers then improved the basic model and used it to develop two models, each of which performed a specific task.

The FAHR-FaceAge model estimates a person's biological age based on a single photo, while the FAHR-FaceSurvival model predicts a cancer patient's chances of survival, by combining facial features and other clinical information. Using the FAHR-FaceSurvival model, researchers found that 52.5% of the 122 patients classified as high risk of death died within 90 days of assessment, consistent with the model's predictions.

On the other hand, the death rate during the same period did not exceed 2% among 975 patients who were classified as the least at risk. When evaluating the FAHR-FaceAge model, researchers found that 21% of 1,060 patients whose biological age was more than 20 years older than their chronological age died within 90 days.

In contrast, the mortality rate was about 6% among 4,195 patients whose biological age was at least ten years younger than their chronological age. Cosmetic changes to the face, such as surgeries, Botox injections and cosmetics, did not affect the performance of the AI ​​models, according to the researchers. 'Proof of a promising concept' The study's authors stress that AI models are not intended to replace doctors or established clinical assessments, but rather describe them as an emerging tool that may, in the future, help assess a patient's health status and make decisions regarding treatment options.

“If we can detect signs of accelerated aging or increased mortality risk indicators at an early stage, we may be able to identify people who need additional testing to detect serious health conditions,” Mack said. Although the new study was limited to cancer patients, Mack noted that further research may reveal whether similar findings apply to healthy people who want to know more about their life expectancy.

Dr. Wilbur Lamm, head of research and professor of pediatrics and biomedical engineering at Emory University, explained that the use of artificial intelligence in the study is interesting, but still needs clinical evaluation. “This is a very important study based on a large data set, and the links we observed between facial images, biological age, and death risk in cancer patients appear strong,” Lam, who was not involved in the study, added in an email.

But proving the actual clinical benefit of this technique requires future studies.” He added: “Moreover, as the researchers themselves point out, it is still unclear whether this tool can be used accurately in specific population groups.

For example, future studies should be conducted to test the extent to which results are affected by facial features, skin color and different age groups, before adopting this technology in clinical practice. Miroslav Popper, Professor of Video Processing and Machine Intelligence at the University of Surrey, said in a statement distributed by the British Center for Scientific Media that the study provides “convincing evidence” of a statistical association between facial appearance and the chances of survival in cancer patients receiving radiotherapy.

But Popper, who was not involved in the study, explained that “predicting results does not mean explaining them.” He added that despite the reference to “biological age,” the AI ​​estimates age based on facial appearance, rather than directly measuring any underlying biological process of aging.

Popper said the researchers linked the facial areas the algorithm relied on to what doctors described as “the appearance of a dying person,” but the study did not test whether the AI ​​was able to spot any indicators that doctors couldn’t. He added that the study did not prove that these predictions provide clinically useful information beyond doctors' assessment of the patient's general health condition and his ability to tolerate treatment, two factors that were not evaluated in the study.

“This is a promising proof of concept, but it is not ready for clinical use,” Popper concluded. “The tool still needs tests that compare its results with the doctors’ own estimates, and evaluate its performance in different hospitals and in different patient populations.”