New AI tool predicts your biological age from a selfie (2025)

Scientists at Mass General Brigham in Boston have developed and carried out some initial testing of an AI tool called FaceAge, an algorithm designed to tell patients’ biological age from a photograph as simple as a selfie ― not how old they are in years, but how old they are in health.

Biological age is considered crucial in helping doctors determine the most appropriate therapy, such as whether a cancer patient is healthy enough to tolerate an aggressive treatment. FaceAge requires more testing before doctors can begin using it routinely, but scientists said that in the next week or two, they expect to begin enrolling about 50 patients in a pilot study.

Researchers said they trained FaceAge using about 59,000 photographs of people ages 60 and older who were presumed to be healthy. Most of the photos were publicly available on Wikipedia and the internet movie database IMDb, while some came from UTKFace, a large-scale dataset with pictures of people from less than a year old to 116 years old.

The developers of FaceAge tested it on a group of 6,200 cancer patients, using photographs taken at the start of radiotherapy treatment. The algorithm determined that when it came to their health, cancer patients were, on average, about five years older than their chronological age. Moreover, the tool found that the older their faces looked, the worse their survival outlook.

Scientists then conducted an experiment in which they asked eight doctors to determine whether terminal cancer patients would be alive in six months based first on the patient’s photograph alone, then on the photograph and clinical information, and finally based on FaceAge and clinical information.

“We found that doctors on average can predict life expectancy with an accuracy that’s only a little better than a coin flip,” when using a photo alone for their analysis, said Raymond Mak, a radiation oncologist at Mass General Brigham and one of the lead investigators for the study.

When using a photo alone, the doctors were right about 61 percent of the time. Given photos and clinical information on the patients, they were right about 74 percent of the time. Provided with FaceAge and medical chart information, the doctors’ accuracy reached 80 percent.

In a news conference last week, Mak said he tested FaceAge using the photograph of a patient whom he first met four years ago, an 86-year-old man with terminal lung cancer.

“Some doctors would hesitate to offer cancer treatment to someone in their late 80s or 90s with the rationale that the patient may die of other causes before the cancer progresses and becomes life-threatening,” Mak said. “But he looked younger than 86 to me, and based on the eyeball test and a host of other factors, I decided to treat him with aggressive radiation therapy.”

Several years later, when Mak ran the man’s photograph through FaceAge, “we found he’s more than 10 years younger than his chronological age.” The patient is now 90, “and still doing great,” Mak said.

Researchers stressed that FaceAge is not intended to replace a doctor’s assessment, but rather to provide an objective measure to help fill out a clinical picture.

“I think it’s really important to know that different people age at different rates, and, as they’re showing here, that clearly seems to have a major effect on their actual prognosis,” said Gregory Schwartz, a scientist at University Health Network’s Princess Margaret Cancer Center in Toronto, who was not involved in the research.

Having a face with a lot of mileage on it, however, does not necessarily doom someone to an early grave.

In a second demonstration of FaceAge, the developers had it analyze photographs of actors Paul Rudd and Wilford Brimley when each man was 50. The algorithm determined that Rudd’s biological age was about 43; Brimley’s was almost 69 (though he would live to 85).

Irbaz Riaz, an assistant professor of medicine and senior associate consultant in the department of AI and informatics at Mayo Clinic, called FaceAge “a promising early-stage tool.”

While the tool does not replace a doctor’s experience, said Riaz, who did not work on the study, “it could standardize the subtle visual assessments we make every day. That said, clinicians will need to understand how the model was trained, when it might be biased and where it could add value without overstepping its role.”

Nasim Eftekhari, vice president of applied AI and advanced analytics at City of Hope cancer treatment and research center in Duarte, California, who did not participate in the study, called the tool “an incremental improvement,” saying, “if this goes through validation and bias testing and approvals, and all of that, this could be an additional biomarker, at best,” to go with cancer stage, characteristics of the tumor and other factors.

Developers of FaceAge acknowledged that should the technology be approved by the Food and Drug Administration, ethical guidelines will need to be established to govern its use and access to its information.

“This technology can do a lot of good, but it could also potentially do some harm,” said Hugo Aerts, director of the Artificial Intelligence in Medicine program at Mass General Brigham and another lead investigator on the FaceAge study.

Aerts said hospitals have “very strong governance committees and regulatory guidelines that they have to adhere to [to] make sure these AI technologies are being used in the right way, really only for the benefit of the patients” and not for others, such as insurers.

Privacy has been a key concern for earlier technologies that attempt to show how we might use age based on a photo, such as FaceApp.

Although initial testing of FaceAge focused on cancer patients, the scientists plan to also measure its performance for other conditions.

Aerts said that the tool still needs to be trained to deal with numerous variables that can affect a photograph of a face: lighting, makeup, skin tone and, of course, our attempts to look younger through plastic surgery.

“So this is something that we are actively investigating and researching,” Aerts said. “We’re now testing in various datasets [to see] how we can make the algorithm robust against this.”

While the tool still has much to learn, we may have something to learn from it, too.

“It is important to know that the algorithm looks at age differently than humans do,” Aerts said. “So, for example, being bald or not, or being gray, is less important in the algorithm than we actually initially thought.”

Mak said the scientists are still trying to figure out what features FaceAge focuses on in the photographs when it estimates biological age.