Can a photo help predict if you’ll survive cancer? Researchers think so

A simple photo may be able to predict cancer outcomes, according to a new study, and researchers are hoping the artificial intelligence being used to do so will be applicable to other diagnostics.

“There’s been a lot of interest in what we think is essentially a new field in medicine,” said Dr. Raymond Mak, a radiation oncologist at Mass General Brigham Cancer Institute and a faculty member in the Artificial Intelligence in Medicine program at Mass General Brigham, to Global News.

Mak is part of a team of researchers who have developed AI models that they say can predict a person’s biological age and chances of surviving cancer treatment using a singular photograph of their face, according to a study published in The Lancet medical journal on Monday.

Researchers trained the models using 40 million photographs including hundreds of thousands with age labels on them. They also had access to more than 30,000 cancer patient photos and outcomes.

“What we’re able to show is that these two distinct health states that we’re able to extract just from a simple face photograph like a selfie, when we applied it to thousands of other cancer patients, were independently very prognostic,” Mak said.

The researchers not only looked at what the outcomes were, but dug deeper into how the model was making the decision.

“These are pretty abstract things. So what we were able to do, there are some technical methods to essentially ask the algorithm what part of the human face is most important for you to make a given prediction,” Mak said.

The age estimation algorithm was looking at the temples, the forehead and the nasolabial fold, Mak said, while the mortality prediction algorithm was looking at the bridge of the nose and the areas under the eyes.

At first it may seem obscure to predict a health outcome as serious as cancer mortality with just a picture of a face, but Mak said it is not that different from what doctors already do every day.

“Doctors are trained from the very beginning of medical school to make subjective assessments of people’s health, to guide decision-making,” he said. “You walk in the room, you eyeball them and you document.”

He said being able to use an algorithmic model could help doctors make these predictions to determine the best course of medical intervention.

What researchers say is new, however, is the ability to run the model on any photograph, including one of a person from years ago, to better plot health history.

“There’s no other test in medicine where you can have a time machine,” Mak said.

Miroslaw Bober, a professor of video processing and machine learning at the University of Surrey who was not a part of the study, said the concept is promising but not ready for clinical use.

“Predicting an outcome is not the same as explaining it. Despite the references to ‘biological age’, the AI estimates age from facial appearance rather than directly measuring any underlying biological ageing process,” Bober said in a statement released by the Science Media Centre.

“The study did not test whether the AI recognizes anything a doctor would not. It has not shown that these predictions add clinically useful information beyond doctors’ assessments of a patient’s general health and fitness, which were not measured in the study.”

But Mak said researchers and clinicians around the world are excited about how the technology could identify new biomarkers to help make earlier diagnoses.

“We think that, if there’s an expert doctor that can look at somebody’s face and make a diagnosis, we think we can train an AI algorithm to match that capability,” he said.

He said the team is currently looking at developing an algorithm that can help predict red blood cell count to help diagnose anemia without or before the use of a blood test.

“From a democratization of healthcare standpoint, anybody in the world with a camera can access these tests eventually,” Mak said.

But the team did have to weigh the ethical implications of training an AI model on facial data in order to get results.

“There’s a privacy issue of did the patient consent or assent to having their data either collected or analyzed, and then what do people do with that information,” Mak said, adding that in the wrong hands a readout could be used to deny medical insurance of further rationing.

“We ultimately decided to move forward with the research, obviously, because we felt like the world needed to know that AI has advanced to the point where it can do these kinds of analyzes just from a simple photograph, so that hopefully we can get ahead of the ethical and regulatory concerns by engaging ethicists, regulatory bodies to put up the right guardrails before people are misusing it,” he said.

Right now, the research is gaining a lot of public traction. Alongside the publication of their research, the team went live with a website where people could upload photos and find out their biological age.

As of Wednesday morning, the site had 50,000 submissions.