Intelligent Stroke Detection Model Leveraging CT Imaging and Deep Convolutional Networks

Abstract

This Brain Stroke Detection System accelerates and enhances the diagnostic process by leveraging a three-layer Convolutional Neural Network (CNN) to analyse non-contrast CT scans. Built with a Python backend supported by Flask and a responsive HTML/CSS/JavaScript frontend, it delivers an intuitive clinical interface. Model Architecture & Performance. The system’s custom CNN demonstrates exceptional performance, achieving 98 % training accuracy and 97 % validation accuracy, validating its strength in identifying stroke-related abnormalities with high precision. Recent comparative studies report similar performance—ensemble CNNs combining InceptionV3, Xception, and MobileNetV2 achieved 98.9 % accuracy, 98.5 % recall, and AUC of 98.7 % on similar stroke CT datasets Dataset & Generalizability The model was trained on a balanced dataset of 2,501 CT images, of which 1,551 are normal and 950 display stroke indications. Such diversity enhances generalizability and minimizes overfitting—reflecting best practices in medical imaging AI Clinical Applicability By integrating a high-accuracy CNN with an accessible web interface, the system offers a scalable, reliable, and clinically viable solution for stroke detection. Its performance aligns with state-of-the-art deep learning tools for CT-based diagnosis, highlighting AI’s transformative role in medical imaging