Within seconds, FaceAge posts the participant’s biological age — an estimate of how old their body appears. Participants can add new images to track how they age over time.
Researchers are using the uploaded photos to investigate which lifestyle habits and diseases are associated with looking younger or older. They stressed that the prediction tool is for research, not diagnosis. More than 10,000 people have signed up since the portal launched quietly in March.
As the database grows, it will help the team test and improve algorithms across larger and more diverse groups. Until now, the tool was limited largely to hospital patients. They’re also collaborating with researchers outside Boston to expand FaceAge.
The study, which tested whether the AI tool could read health signals from ordinary photographs, was published in The Lancet Digital Health on Monday. The researchers first trained a base model, FAHR-Face, on more than 40 million face images. They then fine-tuned it for two tasks: FaceAge estimates biological age and FaceSurvival predicts survival after cancer treatment.
Among 38,000 Harvard patients, those who looked at least five years older than their actual age had a 21 percent higher risk of death, after accounting for age, sex, cancer type, and other factors.
“To our knowledge, this is the first demonstration that an AI model can directly predict cancer survival from a simple photograph,” said Dr. Hugo Aerts, director of the Artificial Intelligence in Medicine program at MGB.
Pratik Shah, a professor of pathology at the University of California, Irvine, who directs its Machine Learning for Health Research Program and was not involved in the study, called it an innovative use of AI.
Shah said estimating biological age from faces is not new. What sets this study apart, he said, is the foundation model trained on millions of images and its large sample of cancer patients.
But he cautioned that the findings come from historical data, not a clinical trial, and that performance varied by cancer type. “The facial-age model showed significant survival associations in 18 of 28 cancer categories,” he said. “The model may not perform consistently across cancer types.”
FaceAge is also being tested in two clinical trials. In one, at Brigham and Women’s Hospital, researchers are enrolling 60 patients ages 65 and older with early-stage lung cancer before radiation. They’re testing whether FaceAge can flag the frail patients who need extra support, such as nutrition or physical therapy, before cancer treatment.
“The care gap we are trying to close is that most older patients with early-stage lung cancer have geriatric vulnerabilities, but very few ever receive a formal geriatric assessment,” said Dr. Raymond Mak, a radiation oncologist at MGB Cancer Institute and a lead author of the study. “If a simple photograph can flag the most vulnerable patients, that would be a practical, scalable way to address this gap.”
The second trial, led by thoracic surgeons at MGB, combines FaceAge with data from wearable devices to flag patients at risk of surgical complications.
“This extends the work beyond radiation therapy and into the surgical setting, where estimating a patient’s physiologic reserve before a major operation is just as important,” Mak added.
Initially, researchers didn’t know what the AI was looking at. But now, attention mapping, a visual representation that looks like a heat map, showed FaceAge focusing on the nasolabial folds, the lines running from the nose to the corners of the mouth, and on the forehead above the eyes and the temples. FaceSurvival looks mainly at the area below the eyes and around the bridge of the nose.
“We don’t know exactly what these algorithms are looking at,” said Mak. “The attention maps tell us where on the face the algorithm is looking to make its prediction, but not precisely which underlying visual features are driving the prediction. Thus, we cannot say the model is measuring crow’s feet, eye bags, wrinkles, pigmentation, skin texture, or temporal muscle wasting individually.”
When the first study came out, people questioned whether makeup, lighting or cosmetic surgery could skew the results. The researchers say they mostly can’t. Pose, lighting, expression, and makeup had little effect, and neither did most plastic surgery procedures. Rhinoplasty resulted in people’s biological age being about one year older, which Mak said could reflect tissue loss from the surgery. He said this suggests that AI isn’t relying on the same cues people do.
A person’s lifestyle leaves major marks. Heavy alcohol use, drug use, and smoking were associated with looking biologically older. Smoking showed the strongest pattern: Current smokers looked older than former smokers, who looked older than people who had never smoked.
“As the population ages, physicians need better ways to estimate physical reserve and frailty before recommending intensive treatments such as radiation, surgery or systemic therapy,” said Mak. “Our goal is to make face photograph-based health assessment more quantitative, consistent and scalable so it can support personalized treatment decisions rather than replace physician judgment.”
Sarah Rahal can be reached at sarah.rahal@globe.com. Follow her on X @SarahRahal_ or Instagram @sarah.rahal.