A Novel Hybrid Model for Brain Ischemic Stroke Detection Using Feature Fusion and Convolutional Block Attention Module

Brain stroke is the second leading cause of death worldwide, following ischemic heart disease. Ischemic stroke occurs when blood vessels are obstructed by a thrombus or other blockages. Prompt and accurate diagnosis of ischemic stroke is critical for patient survival. This study proposes a novel approach for ischemic stroke detection from computed tomography (CT) images, utilizing a hybrid feature extraction technique combined with a convolutional block attention module (CBAM). The hybrid feature extraction leverages the strengths of two pre-trained models, DenseNet121 and MobileNetV3, through feature fusion to provide a comprehensive representation of brain CT images. The CBAM module is integrated to enhance the most relevant features by focusing on both channel and spatial attention mechanisms, significantly improving the model’s ability to detect ischemic strokes with high accuracy. The proposed approach was developed and evaluated on a unique first-hand dataset (Dataset 1) collected from the Specialized private Hospital in Palestine. To further demonstrate the robustness and generalizability of the method, it was also tested on a public dataset (Dataset 2). The results show that the proposed model achieved outstanding performance across all metrics: on Dataset 1, it reached an accuracy of 99.21%, precision of 99.17%, recall of 99.32%, and an F1-score of 99.24%. On Dataset 2, it achieved an accuracy of 98.73%, precision of 98.71%, recall of 98.74%, and an F1-score of 98.88%. The exceptional results are attributed to the power of feature fusion, which creates a robust and comprehensive feature representation. In addition to the CBAM, which refines the feature maps by selectively focusing on the most important channels and spatial regions in brain CT images. This has led to substantial improvement in the detection of ischemic stroke areas.