Abstract

Cerebrovascular pathologies, particularly strokes, are a leading cause of global mortality and morbidity, where timely diagnosis is critical for effective clinical intervention. This research introduces an advanced diagnostic framework for early stroke identification by integrating Computed Tomography (CT) imaging with a hybrid computational approach. The system utilizes a Genetic Algorithm (GA) to optimize feature selection from neural network outputs, identifying the most discriminative biomarkers within the image data. These refined features are subsequently classified using a Bidirectional Long Short-Term Memory (BiLSTM) network. Rigorous validation through cross-validation showed that the proposed model achieved a 96.5% accuracy, significantly outperforming traditional classifiers like Random Forests and Support Vector Machines. By utilizing metrics such as AUC-ROC and F1-score, the study demonstrates that this architecture serves as a robust decision-support tool, enabling medical practitioners to implement more precise and rapid stroke management strategies.