Acute Ischemic Stroke Detection on Non-Contrast CT: A Deep Learning Approach

Acute ischemic stroke is a leading cause of disability and death, and effective treatment requires quick and accurate diagnosis. Non-contrast CT (NCCT) is widely used for initial screening of ischemic stroke, but detection is challenging due to subtle or indistinguishable changes on NCCT. Researchers trained a deep learning algorithm to classify patients with acute ischemic stroke and segment stroke lesions using hyperacute NCCT images and diffusion-weighted MRI as the reference standard. The ResNet50 model achieved the best performance with 98.5% accuracy, 97.4% precision, and 100% recall on the evaluation set. Classification performance remained strong even for lesions smaller than 5 mL, which comprised the majority of evaluation cases. For segmentation using U-Net architectures, performance was acceptable for large lesions but declined significantly for smaller lesions. These findings demonstrate the feasibility of deep learning for acute ischemic stroke detection and represent a step toward faster triage and treatment for stroke patients.