Stroke Detection Using Deep Learning

Stroke is a critical medical condition that requires rapid and accurate diagnosis to improve patient recovery out-comes. This study presents a deep learning-based approach for stroke detection using Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks. Initially, CNN and BiLSTM models are evaluated separately before integrating them to enhance feature extraction. The optimized model leverages EfficientNetB0 as a pretrained CNN backbone, combined with additional layers such as BiLSTM, Flatten, Dense, and Dropout layers to improve classification accuracy. Achieving a training accuracy of 98 percent and a validation accuracy of 95 percent, the model demonstrates high effectiveness in stroke diagnosis. EfficientNetB0 extracts intricate features from CT scan images, while the BiLSTM network processes spatial and sequential dependencies, leading to more reliable classification. The system categorizes strokes into three types: hemorrhagic, ischemic, and silent stroke. For practical application, the model is integrated into a hospital management platform, allowing doctors and patients to upload CT scans for instant analysis, thus facilitating timely medical intervention. The findings suggest that combining EfficientNetB0 with BiLSTM provides a robust and efficient solution for automated stroke detection, making it a valuable tool in clinical settings.