New AI instrument estimates biological age by analyzing a face

A novel artificial intelligence instrument crafted by researchers at Mass General Brigham and Harvard Medical School utilizes a photograph of a patient’s face to forecast biological age and cancer survival duration, insights that physicians can employ to customize treatments.

“We all recognize that individuals age uniquely. A person’s chronological age is determined by their birth date, but that does not equate to biological age, which serves as an actual indicator of their physiological wellness and lifespan,” stated Hugo Aerts, the study’s co-senior author, director of MGB’s Artificial Intelligence in Medicine initiative, and professor of radiation oncology at HMS. “A person’s biological age is influenced by various factors, including lifestyle, genetics, and other health aspects. We had this notion that a person’s appearance could indeed be a reflection of their biological age.”

Guided by scientists at MGB’s Artificial Intelligence in Medicine Program, the researchers educated FaceAge, their deep-learning algorithm, on over 58,000 images of healthy individuals with known ages and more than 6,000 images of cancer patients whose ages and clinical outcomes were established.

The algorithm revealed that cancer patients’ FaceAge was, on average, five years older than their chronological age. It also discovered that appearing older correlated with poorer outcomes for individuals battling various cancer types.

“We had this idea that how old a person looks could actually be a reflection of their biological age.”

Hugo Aerts

Evaluating health based on appearance is not an unfamiliar concept, Aerts noted. Physicians routinely perform a visual appraisal — which Aerts referred to as the “eyeball test” — when they enter the room. This assessment can include observations such as whether the patient is utilizing a wheelchair, their physical robustness, and any obvious illness.

Research indicated, however, that the eyeball test — at least when conducted by human doctors — is not a reliable predictor of short-term life expectancy.

Published in the journal The Lancet Digital Health in early May, the study, funded by the National Institutes of Health, requested 10 clinicians and researchers to estimate short-term life expectancy using images of 100 terminal patients who were undergoing palliative radiation therapy.

On average, they performed only marginally better than random chance, even with knowledge of the patient’s chronological age and cancer status. Prediction improved, however, when clinicians received FaceAge information for those patients.

Raymond Mak, a faculty member at the Artificial Intelligence in Medicine Program, HMS associate professor of radiology oncology, and co-senior author of the study, stated that having a clearer understanding of a patient’s biological age and their expected remaining time allows oncologists to better customize treatments.

He described a lung cancer patient who, while chronologically 86, appeared significantly younger. This influenced Mak’s recommendation for a more aggressive treatment. Today, the individual continues to thrive at age 90. When Mak utilized FaceAge to analyze a picture of the patient at the treatment time, the algorithm assessed his biological age as 10 years younger than his chronological age.

The reverse can also be true, Mak noted, as patients who are more delicate than their chronological age may imply could require less intensive treatment because of what their body can endure.

“We hypothesize that FaceAge could function as a biomarker in oncology to quantify a patient’s biological age and assist a physician in making these difficult decisions,” Mak stated.

FaceAge has proven effective across various cancer types, Mak and Aert mentioned, and they are investigating its potential usefulness in predicting outcomes in different diseases.

The algorithm operates using deep learning, meaning that it adapts as researchers instruct it on thousands of images of individuals with known outcomes.

The algorithm revealed that cancer patients’ FaceAge averaged five years older than their chronological age.

Researchers, however, are not aware of which precise cues catch FaceAge’s attention, Aert mentioned. It’s likely that the algorithm is focusing on aspects different from what a doctor might consider, such as wrinkles, gray hair, and baldness. If that holds true, it would be particularly advantageous, he said, because it offers a new perspective to doctors’ evaluations of a patient’s condition.

Aerts and Mak stated that FaceAge would not be used independently to determine treatment courses but would act as a resource available to physicians. It could contribute not only to establishing initial treatment but also to monitoring changes over time, notifying a doctor if a patient seems to be deteriorating.

Before it is implemented in clinical settings, however, further testing on diverse patient populations is necessary.

“In the clinic, the impact could be substantial, as we now possess a method to continuously monitor a patient’s health status — before, during, and after treatment — which could aid us in better predicting the risk of complications post, for example, major surgery or other therapies,” Aerts concluded.