Non-inferiority of deep learning ischemic stroke segmentation on non-contrast CT within 16-hours compared to expert neuroradiologists
Abstract
Acute ischemic stroke is a leading cause of death and disability worldwide, requiring rapid diagnosis and treatment. Automated segmentation of ischemic stroke lesions on non-contrast CT scans could accelerate clinical decision-making. However, clinical adoption requires demonstrating that AI performance is non-inferior to expert radiologists.
We conducted a non-inferiority study comparing deep learning-based stroke segmentation to expert neuroradiologists on non-contrast CT scans acquired within 16 hours of stroke onset. Our deep learning model was trained on a large dataset of stroke cases and evaluated against expert annotations from board-certified neuroradiologists.
Results demonstrate that the deep learning approach achieves non-inferior performance compared to expert neuroradiologists for acute ischemic stroke segmentation. The model shows consistent performance across different imaging protocols and patient populations, suggesting its potential for clinical deployment to assist in rapid stroke diagnosis and treatment planning.