AI Framework for Stroke Detection and Classification Using Machine Learning

Stroke is a major health problem that happens when the blood supply to the brain is interrupted. It may cause brain cell damage, long-term disability, or even death. Around the world, it is still one of the main causes of death and disability and there is a need for quicker and more reliable ways to diagnose it especially in countries with limited resources. This research proposes an AI framework that combines stroke classification and lesion detection in real time using CT scans. EfficientNet-B7 was trained on 13,200 images for classification into normal, ischemic, and hemorrhagic types, while YOLOv10 was trained on 13,695 annotated images with masks and labels for lesion localization. The two models were integrated into a single diagnostic system, where EfficientNet-B7 achieved 99% accuracy, and YOLOv10 reached a mean Average Precision of 96.3% at a confidence level of 0.4. A web interface was developed that allows clinicians to upload CT images and receive instant results. The system was also tested on real CT images from the Sana’a Radiology Center and achieved 100% accuracy on this limited dataset, showing its ability to perform in real conditions. Unlike earlier studies that treated classification and detection separately or lacked clinical readiness, this work provides a practical, unified tool that is suitable for clinical use.