AI-STtroke, an innovative start-up based in Montpellier, has taken a major step in its development by creating a vast game data for stroke pre-diagnosis. In collaboration with the Nîmes University Hospital, the company filmed 300 patients in the acute phase of stroke, thus generating 20,000 videos and 6 million images. These resources allow its artificial intelligence to train and obtain promising results in detecting strokes. This project illustrates AI-Stroke’s commitment to using real data to train its AI, thereby increasing diagnostic accuracy.
Strokes represent the second cause of death worldwide, affecting millions of people each year. Victims lose on average 2 million neurons every minute in the event of an accident. According to a study, patients who arrive at the emergency room within three hours of the first symptoms appearing have less disability three months after the stroke. These statistics highlight the critical importance of early detection and rapid intervention, which can significantly improve patient outcomes.
Challenges to overcome for early detection
The urgency for rapid intervention is obvious, but the reality presents many obstacles. Many strokes occur outside of hospitals, which complicates initial care. Indeed, the variability of symptoms from one patient to another makes diagnosis difficult. Firefighters and emergency workers, who are often the first to respond, do not always have the necessary tools to effectively diagnose strokes. Almost 40% cases of stroke escape first aid, and an American study found that 22% of strokes are not detected during emergency room visits.
“Taking care of a patient within the clinical window of care – a maximum of 4 to 6 hours after the first symptoms – is a real challenge,” underlines Eric Thouvenot, university professor and hospital practitioner (PUPH) in neurology at Nîmes University Hospital. “We are convinced that the solution currently developed by AI-STROKE will provide a satisfactory response to one of the major challenges in stroke care: responsiveness. »
A digital neurologist accessible to everyone
AI-Stroke has designed artificial intelligence that works like a digital neurologistallowing a pre-diagnosis before formal examination by a doctor. During the first assessment, the neurologist asks the patient to perform a few simple exercises, such as “smile big,” “raise both arms,” or “repeat these few phrases.” These gestures make it possible to check the presence of the main characteristic symptoms of stroke, in particular mouth deformity, weakness on one side of the body and speech problems.
To establish a diagnosis with comparable precision, even when a neurologist is not available nearby, AI-Stroke has developed artificial intelligence that can be integrated into a tablet or smartphone. During the assessment, the AI films the patient while they perform these exercises, then analyzes the videos by comparing them to the 6 millions d’images already collected during the creation of the dataset. The AI is then able to indicate whether or not the patient’s reactions are consistent with a stroke, providing a crucial first assessment.
The long-term goal is to integrate this technology into tablets used by the firefighters during their interventions. This would communicate essential information to hospitals, facilitating rapid referral of patients to one of the 150 centers specialized in stroke present in the territory. This innovation could thus reduce the precious time lost before the neurologist’s final diagnosis.
“Now that we have a very substantial dataset, we can focus all our efforts on training our AI, the first results of which are more than encouraging,” says Cédric Javault, CEO of AI-Stroke.
AI can already diagnose a stroke in 74% of casesand performance continues to improve with each new data integrated. The more data AI-Stroke manages to collect, the more the accuracy will increase, until it becomes as precise as an expert neurologist.
Towards clinical trials and marketing
AI-Stroke plans to launch first clinical trials in 2025a crucial step before the planned marketing of fin 2026. Integrating AI into the stroke detection process could revolutionize the way strokes are diagnosed and treated, providing a rapid and effective response to stroke victims.
The start-up is thus positioning itself as a key player in the fight against stroke, combining technological innovation and public health imperatives. The results obtained to date reinforce the hope of early detection and better management of strokes, helping to reduce neurological after-effects for patients.
