Cerebrovascular Accident Prediction using Deep Learning Techniques

Abstract Stroke or cerebrovascular accident (CVA) is a major cause of mortality and chronic disability in the entire world with an early and precise detection being a key to saving lives of patients. The provided work offers a medical image classification system based on deep learning to predict stroke in real-time with the help of convolutional neural networks (CNNs). The proposed model is trained on the medical images of the brain classified into normal and stroke and the images are resized to 256 x256 and normalized to learn the features with strength. The CNN model is made up of successive convolution, max-pooling, flattening and sigmoid-activated dense layers that are optimised through Adam optimiser with binary cross entropy loss. A batch size of 32 and 10 training epochs with an 80:20 training -validation split are used. To achieve the real-time clinical usability, the trained model is launched with the help of the web-based application with Streamlit, which facilitates the interactive upload of medical images, stroke prediction based on probabilities, choosing specific decision thresholds, and visualizing the confidence levels with bar charts. The created framework shows that it is possible to incorporate the stroke prediction system based on deep learning into real-time medical decision support systems and improve diagnostic support.