Brain Stroke Detection Using Deep Learning:An AI-Based Medical Image Analysis System

Abstract

Brain stroke is the second leading cause of death globally, demanding rapid and accurate diagnosis from CT scans.
This paper presents a custom Convolutional Neural Network (CNN) comprising four convolutional layers
(1→32→64→128→256 channels) with batch normalization, max pooling, and dropout (p = 0.5), implemented in
PyTorch. The model processes 128×128 grayscale CT images and performs binary classification (Stroke / Normal)
via sigmoid activation with 100% test accuracy. Comparative evaluation against Random Forest (99%), SVM (97%),
and Logistic Regression (95.5%) validates the superiority of end-to-end deep learning for medical image analysis.
The system is deployed as a Flask web application with drag-and-drop image upload, real-time prediction, scan
history, Chart.js analytics dashboard, and Docker containerization.