Advanced Deep Learning for Stroke Classification Using Multi-Slice CT Image Analysis

Brain stroke is a leading cause of mortality and disability globally, necessitating rapid and accurate diagnosis for timely intervention. While Computed Tomography (CT) imaging is the gold standard for stroke detection, manual interpretation is time-consuming, prone to error, and subject to inter-observer variability. Although deep learning models have shown promise in automating stroke detection, many rely on 2D analysis, ignore 3D spatial relationships, or require labour-intensive slice-level annotations, which limits their scalability and clinical applicability. To address these challenges, we propose MedHybridNet, a novel hybrid deep learning architecture that integrates convolutional neural networks (CNNs) for local feature extraction with Transformer-based modules to model global contextual dependencies across volumetric brain scans. Our main contribution is twofold: (1) the SliceAttention mechanism, which dynamically identifies diagnostically relevant slices using only patient-level labels, eliminating the need for costly slice-level annotations while enhancing interpretability through attention maps and Grad-CAM visualizations; and (2) a cGAN-based augmentation strategy that generates high-quality, pathology-informed synthetic CT slices to overcome data scarcity and class imbalance. The framework processes complete 3D brain volumes, leveraging both CNNs and Transformers in a dual-path design, and incorporates hierarchical attention for refined feature selection and classification. Evaluated via patient-wise 5-fold cross-validation on a real-world dataset of 2501 CT scans from 82 patients, MedHybridNet achieves an accuracy of 98.31%, outperforming existing methods under weak supervision. These results demonstrate its robustness, generalization capability, and superior interpretability. By combining architectural innovation with clinically relevant design choices, MedHybridNet advances the integration of Artificial Intelligence (AI) into real-world stroke care, offering a scalable, accurate, and explainable solution that can significantly improve diagnostic efficiency and patient outcomes in routine clinical practice.