High blood pressure and diabetes: how artificial intelligence detects them using a video of the face

In previous experiments, using a 30-second recording, the system had identified high blood pressure with 95 per cent accuracy. But the most interesting thing is that even when the video clip was drastically shortened to 5 seconds, the algorithm continued to work, providing diagnoses with 90.3 per cent accuracy. It is as if the machine is able to detect, via the face, a sort of cardiovascular ‘signature’ associated with high blood pressure. But there is more. The new study set out to verify whether the same approach could also work for diabetes.

Facial blood flow patterns play a key role here. Using a 30-second recording, the algorithm identified diabetes with 88.2 per cent accuracy; when the video was shortened to just five seconds, the accuracy was 81.2 per cent.

These are very interesting figures, particularly when considering their potential use in large-scale screening.

This is because high blood pressure and diabetes share a particularly insidious characteristic: they can be present for years without causing any obvious symptoms. They are two silent killers which, unseen, damage the arteries, heart, brain, kidneys and other organs, increasing the risk of a heart attack and stroke.

Worldwide, it is estimated that around 1.4 billion adults aged between 30 and 79 live with high blood pressure, whilst there are around 589 million people with diabetes. However, there is a significant number of undiagnosed cases. That is why this idea of mass screening using a short video could prove to be particularly promising.

The future of screening could, in fact, involve a camera installed in a chemist’s, in a doctor’s waiting room, in a shopping centre – or perhaps even built into a smartphone.

You stand in front of the camera for a few seconds, then the artificial intelligence analyses the footage and provides a recommendation: you may be at risk, so you should check your blood pressure or have the necessary tests carried out.

This would not be a fully substantiated diagnosis (or at least ‘not yet’). Rather, it would be a warning sign, capable of bringing to the fore people who would normally slip through the net of traditional screening. And this is precisely where the real promise of the technology lies: not in replacing the doctor, but in being able to identify at an earlier stage those who do not yet know they have a problem.

Before we start imagining millions of people being diagnosed via a selfie, however, we need to keep our enthusiasm in check. The study was carried out on just 215 participants at a single centre. The researchers themselves emphasise that the model will need to be validated on much larger groups and, above all, on more diverse populations. But that’s not all.

The algorithm also attempted to estimate blood pressure directly without using a conventional monitor, but the results here were less convincing. The variability in the error was still too high. In short, it is not yet time to consign the sphygmomanometer to the attic. But that is certainly not the aim of this study.

If these findings are confirmed, the real revolution could be something else entirely. At present, to find out if we have high blood pressure, we have to measure it. To find out if we have diabetes, we have to undergo specific blood tests. And many people do not do these, because they ‘feel’ fine. A screening system of this kind, on the other hand, could identify people before they have a reason to see a doctor.

This is the most interesting aspect of the Japanese research: turning a mundane action, such as standing in front of a camera for a few seconds, into a potential opportunity for prevention. Of course, the system will first need to be validated through far more extensive studies involving different populations. But the direction seems to have been set.

The next time you have your blood pressure checked, it might no longer start with a cuff inflating around your arm, but with a camera scanning your face for five seconds.

And the future of artificial intelligence may not just be about making us work less or better. It may be about realising, before we do (and despite the fact that we ‘feel fine’), that something is wrong with our bodies.