A simple selfie might reveal unexpected insights about your health and longevity, particularly for cancer patients.
Researchers from Mass General Brigham in Boston claim they have created AI models that can estimate a cancer patient’s biological age and predict survival chances using just a single facial photograph.
This research, published in the journal The Lancet Digital Health, indicates that analyzing facial features could serve as an additional resource for doctors assessing patient survival and treatment strategies for cancer.
Dr. Raymond Mak, a radiation oncologist at Mass General Brigham Cancer Institute, mentioned that he was quite taken aback to discover that mortality risk signs could be identified from a mere facial photograph.
The AI models, which were trained on over 40 million images, evaluate various facial characteristics to determine a cancer patient’s health and biological age. Biological age, unlike chronological age—which simply counts the years lived—reflects the body’s cellular and physical aging.
Mak elaborated that the age estimation algorithm focuses on areas such as the temples, forehead, above the eyes, and the nasolabial fold, which connects the nose to the mouth’s corners.
Conversely, he noted, the mortality risk prediction algorithm examines entirely different facial regions, like the bridge of the nose and under the eyes.
Interestingly, the facial areas assessed by the AI for biological age and survival risk echo descriptions made by Hippocrates in the Book of Prognostics around 400 B.C. He linked features like a sharp nose, sunken eyes, and sunken temples to muscle wasting and mortality risk.
Mak expressed that realizing their algorithms might reflect Hippocratic wisdom from so long ago was quite astonishing.
However, Mak pointed out that the AI had no knowledge of ancient texts; the algorithms were strictly trained on facial photographs alongside data on patient mortality and age.
He added that these algorithms do not incorporate internet knowledge or wisdom. The eerie feeling of echoing Hippocrates over millennia was notable.
Initially, Mak and his team developed a comprehensive AI model called FAHR-Face (Foundation Artificial Intelligence for Health Recognition-Face), trained on a diverse set of over 40 million facial images.
This model was then refined to create two additional models for specific tasks: the FAHR-FaceAge model estimates biological age from a single photo, while the FAHR-FaceSurvival model predicts cancer survival chances by integrating facial features with other clinical data.
Using the FAHR-FaceSurvival model, researchers observed that among 122 cancer patients flagged as having a high death risk, over half—52.5%—died within 90 days, which was consistent with predictions. In contrast, around 2% of the 975 patients considered low-risk died within the same period.
For the FAHR-FaceAge model, it was noted that 21% of 1,060 patients with biological ages exceeding their chronological ages by over 20 years died within 90 days, whereas only about 6% of 4,195 patients with biological ages at least 10 years younger passed away in that time frame.
Surprisingly, cosmetic modifications to the face, such as surgeries, Botox, or makeup, did not influence the AI’s accuracy.
The authors of the study clarified that their AI models do not aim to replace physicians or current evaluations. Instead, they view this technology as a potential new resource for assisting doctors in assessing patient health and making treatment decisions.
Mak expressed that identifying accelerated aging or higher mortality risks early could prompt screenings for patients who might benefit from further testing for serious conditions.
While the study focused exclusively on cancer patients, Mak mentioned more research is needed to see if similar conclusions could apply to otherwise healthy individuals interested in longevity.
The use of AI in this study is intriguing yet requires clinical evaluation, noted Dr. Wilbur Lam from Emory University.
He described the study as compelling, based on a large dataset linking facial images with biological age and cancer survival risk, though he emphasized that the actual clinical utility needs to be confirmed through prospective studies.
Furthermore, he pointed out the necessity to assess whether this tool works accurately across various subpopulations with different facial features, skin tones, and age groups.
The research presents convincing evidence linking facial appearance to survival in cancer patients undergoing radiotherapy, according to Miroslaw Bober from the University of Surrey.
However, he cautioned that predicting outcomes is different from explaining them. The AI estimates age based on facial appearance rather than directly gauging underlying biological aging processes.
Bober noted the authors associate AI-analyzed facial regions with the “look of dying” that doctors recognize, yet the study did not verify whether the AI recognizes anything beyond a doctor’s assessment.
He remarked that the study hasn’t demonstrated that these predictions provide clinically valuable information that surpasses existing assessments of a patient’s overall health, which were not evaluated in this study. It’s a promising concept, but the technology needs more testing against doctors’ judgements across various hospitals and patient demographics.
