People who have suffered a CVA (cerebrovascular accident) are often taken care of late because of a lack of knowledge of the symptoms. The Montpellier company AI-Stroke is developing software that uses artificial intelligence to detect strokes quickly. The project is in its development phase and should be used by firefighters by the end of 2023. Interview with the founder of AI-Stroke, Cédric Javault.
You want to use artificial intelligence to quickly detect a stroke. Is it because the human is not efficient enough and quick enough to spot it?
At present, a stroke is difficult to spot when you are not a health professional. The clinical signs to realize this are not simple. Indeed, artificial intelligence software, ultimately a digital neurologist, can be very useful and save a lot of time.
What are the signs of a stroke?
There are three main ones. The first is facial paralysis. You have a crooked mouth, for example. The second is that you can no longer lift an arm or a leg. And the third, you have real speech difficulties. And all this suddenly appears.
How will this software work?
If you feel unwell, for example a stroke, you will probably call the fire department. Firefighters are emergency professionals, but they are not specifically trained in strokes. Unfortunately, they make a lot of mistakes.
So the idea is to give firefighters that tool. They will film you, asking you a few questions. You will answer these few questions, we will record and the artificial intelligence will analyze and be able to say if it is a stroke.
We’ll save several minutes. In particular because we will be able to say afterwards to the regulator of the Samu, even to the neurologist on duty if it is really a CVA. If it was checked by the machine. You should know that during a stroke, you lose 2 million neurons per second.
You are in full development of this software. You are in the data collection phase. Concretely, how do you build this software?
Training artificial intelligence requires data. Today, there are no video images of stroke patients. We obtained the agreement of the ethics committee to precisely collect the videos of 300 patients who are having a stroke in the department of Professor Thouvenot in Nîmes.
Concretely, we will offer patients who agree to do a standard test. The one that allows neurologists to diagnose a stroke. And we’re going to film it. We’re also going to do this with people who don’t have a stroke so that AI, artificial intelligence, has both videos of people who are sick and others who aren’t sick and learn gradually to make a difference.
Interview by Morgane Guiomard
Pictured: AI-Stroke, software for rapidly detecting strokes developed in Montpellier. – Credits: AI-Stroke