New AI tool examines selfies to predict cancer survival: Can it replace doctors?

A team of scientists has created a machine learning algorithm that, through a single picture of a patient's face, can predict a patient's survival rate if they have cancer.

The technology, developed by researchers at Mass General Brigham in the United States, was trained using more than 40 million facial images and tested using photographs and medical data from more than 30,000 cancer patients.

What is it about a human face that AI can spot that doctors can miss? As per the study published in The Lancet Digital Health, the system analyzes very subtle facial characteristics that could indicate one's physical condition and biological age.

To address this question, two tools were created: FAHR-FaceAge and FAHR-FaceSurvival.

The first is an estimation of how old a person's body looks, rather than just taking their age based on their date of birth. The second is more advanced, looking at facial characteristics to predict the survival rate of those who have already been diagnosed with cancer.

Interestingly, researchers found that the area beneath the eyes and around the bridge of the nose played an important role in the survival predictions.

The study also found that smoking, heavy alcohol consumption and drug use were associated with signs of accelerated facial ageing. But the technology does not diagnose cancer in a person or determine the exact time when a person might die. Rather, it looks for patterns in groups of patients that correlate with different outcomes of survival.

The device could one day be used to determine whether patients are physically robust enough to receive demanding treatments.

There's one important question that's been raised from independent experts: has the technology offered anything that the doctors cannot get from a medical examination?

The researchers state that additional testing is required before the system could be used in daily medical practice in hospital settings.