Aug. 23, 2024 — A team led by researchers at Yale School of Medicine has created a deep learning model that can accurately project post-surgical outcomes for some patients with large vessel occlusion (LVO) stroke using computed tomography angiography (CTA) scans taken at admission.
"The deep learning model developed by our research team is the first step toward intelligent machinization of stroke neuroimaging protocol," said Sam Payabvash, M.D., Associate Professor of Radiology and Biomedical Imaging and senior author of the study.
Using data from patients undergoing thrombectomy from 2014 to 2020, the Yale team trained three different models on admission CTA scans with and without data including time to surgery, age, sex and NIH stroke scale score. Researchers say the new tool would enable quick and accurate decision-making, with an established 'treatment trigger' to activate the treatment chain after surgery.
"We developed an end-to-end fully automated deep learning model that can predict stroke outcomes from readily available admission brain images and treatment success scenarios, with 78% accuracy in independent validation," Payabvash said.
"It’s worth noting that the model can solely rely on CT angiography scans of the brain, which are invariably present at the time of stroke diagnosis," he added. "Therefore, our model based on imaging information can provide rapid, objective predictions regardless of local expertise and other variabilities, guiding treatment in resource challenged communities.”
March 12, 2026 
