Brain Stroke Detection Using Deep Learning

Abstract

Introduction: Brain stroke represents a critical scientific situation requiring speedy and correct prognosis to mitigate the chance of permanent neurological damage or loss of life. traditional diagnostic techniques, predominantly reliant at the guide interpretation of computed tomography (CT) scans by way of healthcare specialists, are frequently restricted through time delays, inter-observer variability, and the inherent barriers of human judgment. To deal with these demanding situations, this study introduces an automatic stroke detection gadget based totally on a hybrid deep mastering architecture that synergizes Convolutional Neural Networks (CNNs) and long quick-term reminiscence (LSTM) networks. The CNN module effectively extracts spatial features indicative of stroke-associated anomalies, at the same time as the LSTM issue captures temporal patterns across sequential picture slices, improving sensitivity to early and innovative stroke signs. The model is trained on a dataset comprising 2,501 CT pictures, similarly representing stroke and non-stroke instances to hold magnificence balance. In performance evaluation, the CNN and LSTM modules for my part completed validation accuracies of 98% and ninety five%, respectively, whilst the incorporated system yielded a high place under the Curve (AUC) of 0.952, underscoring its robust category functionality. To facilitate scientific adoption, the device is deployed through an internet-based totally utility the usage of the Flask framework, allowing real-time inference via an intuitive physician interface. This platform empowers clinicians to add CT pics and obtain immediate diagnostic feedback, assisting time-sensitive medical decision-making. normal, the proposed framework offers a scalable, accurate, and efficient answer for reinforcing stroke prognosis and affected person care.