Scientists trained AI model using brain scans from 800 stroke patients with known stroke timings
Scientists have developed a new artificial intelligence software that can determine the timing of a stroke and whether treatment is feasible with twice the accuracy compared to conventional methods.
Researchers from Imperial College London in the UK, the Technical University of Munich in Germany and the University of Edinburgh in Scotland collaborated on the software.
They trained the model using brain scans from 800 stroke patients with known stroke timings.
The model was tested on data from approximately 2,000 patients, demonstrating that it outperformed standard visual assessment methods by providing twice as accurate results in determining stroke timing.
The software automatically identifies relevant brain areas in scans and analyzes lesions to estimate when the stroke occurred.
Researchers highlighted that the software's ability to incorporate tissue characteristics and account for variations in lesions contributed to its superior performance compared to traditional scanning methods.
Dr. Paul Bentley, one of the authors of the study, explained that most strokes caused by blood clots are treatable with medical or surgical interventions within 4.5 hours of an onset.
He emphasized that conventional treatments are effective only in the early stages of a stroke; otherwise, they can lead to secondary damage.
Having the information at their fingertips helps doctors make critical decisions about which treatments to apply to stroke patients, said Bentley.
Researchers want to facilitate faster and more accurate treatment for patients using the new AI technology.
The findings were published in Nature Digital Medicine.
