Researchers at the German Center for Neurodegenerative Diseases (DZNE) and the University Hospital Bonn are developing an AI-based computer model to support doctors in stroke treatment. The goal is a digital assistance system that predicts the health status after a mechanical thrombectomy and detects complications to optimize therapy decisions. A feasibility study is examining whether this is possible with data from the "German Stroke Registry" and brain imaging. The project, funded with 250,000 euros by the Helmholtz Association, uses the innovative AI technology of "swarm learning" and involves the CISPA Helmholtz Center for Information Security.
An ischemic stroke, often caused by blood clots, leads to neurological deficits such as paralysis or speech disorders. Without rapid treatment, millions of brain cells die. Therapy options such as drug-induced clot lysis or minimally invasive thrombectomy require quick, individual decisions. The planned AI model is intended to provide predictions about treatment success and risks, with the "explainability" of the AI ensuring that doctors can understand the basis of the prognoses.
The AI is trained with data from the "German Stroke Registry," which records thousands of stroke cases from over 20 clinics, and additional MRI or CT scans. "Swarm learning" enables the analysis of distributed data without central collection. The algorithm "travels" to the data, which remains local, ensuring data protection and efficiency. This creates a model that utilizes the knowledge of all network partners and continuously evolves.
The project starts with three clinics, including Bonn, and initially simulates a swarm in the DZNE computing center. In the long term, a nationwide network is to be established, with plans for international collaborations, for example with Great Britain. The study lays the foundation for more precise, data protection-compliant stroke treatment.
