AI has learned to determine biological age by face

Aging can now be seen not only in the mirror, but also on the screen – if a neural network is involved. Specialists from Massachusetts General Hospital (Mass General Brigham) have developed the FaceAge algorithm, which is able to determine a person’s biological age from facial images and even predict how long he or she will live with cancer.

Interestingly, patients with cancer looked on average five years older than their actual age according to the algorithm. The more FaceAge “aged” a person, the worse their survival prognosis was – regardless of the type of cancer or passport age.

At the start of their work, the researchers used nearly 59,000 photos of healthy people to train the system to recognize features associated with age-related changes. The neural network was then tested on 6,200 cancer patients whose photos were taken at the start of radiation therapy. The AI coped with the task no worse, and sometimes even better than experienced doctors.

When a doctor needs to assess a patient’s condition, he or she often relies on the impression of appearance. But this approach can be subjective. FaceAge offers a more objective and digital way of assessment. What’s particularly valuable is that AI has proven useful in the most difficult cases – for example, in predicting the remaining time to live in patients in the palliative stage.

An interesting experiment involved 10 physicians who were asked to estimate the life expectancy of 100 patients using photos and baseline information. Their results on average were no better than flipping a coin. But when they added the biological age data from FaceAge, the accuracy increased dramatically.

Now the developers are continuing to test the system. In the near future, they plan to expand the databases, take into account different stages of disease, assess the impact of cosmetics and plastic surgery on the accuracy of AI.

According to Professor Ray Mack, one of the authors of the project, FaceAge may become not just a tool for oncology, but a universal indicator of aging and health. He believes that in the future this technology could be used for early detection of age-related diseases and making personalized medical forecasts.

AI is increasingly becoming not only a physician’s assistant, but also a potential diagnostician. And perhaps in the coming years, selfies will be able to tell more about a person’s health than a whole folder of tests.