Automated Deep Learning System for Early Detection of Stroke Using CNN Architecture

Strokes are a major cause of disability worldwide, with ischemic and hemorrhagic strokes accounting for the majority of cases. In India, stroke remains the second most common cause of mortality. Approximately $\mathbf{7 5}-\mathbf{8 5 \%}$ of reported strokes are ischemic, while $15-25 \%$ are hemorrhagic. The proposed system focuses on developing an automated deep learning system for stroke detection using a Convolutional Neural Network (CNN) architecture. The CNN model extracts spatial features from 2,311 CT images of the human brain, sourced from Kaggle. To improve the model's robustness, data augmentation techniques, such as rescaling, shear range, zoom-in, and horizontal flipping, are applied. The system integrates the Adam optimizer to enhance training efficiency and employs the ReLU activation function to ensure better model convergence and accuracy. The CNN model, implemented and tested achieves an accuracy of $89.58 \%$, highlighting its potential for fast and non-invasive stroke detection. By addressing challenges like human error and time delays in conventional diagnostic methods, this system provides improved precision while reducing the manual workload. This enhancement aims to create a scalable and efficient solution that advances stroke care by enabling timely medical interventions, particularly in resource-constrained settings.