A hybrid CNN-transformer model for multi-marker detection in cerebrovascular disease using multimodal imaging

Cerebrovascular diseases (CVDs) demand early detection of multiple lesion markers for proper treatment. This study proposes a hybrid deep learning model combining Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) to capture both local and global features from multimodal MRI sequences (T1, FLAIR, DWI). The model leverages cross-modal attention for fusion and SHAP for clinical transparency. Using the ISLES 2022 dataset, the proposed model achieved 89.5% accuracy, 86.9% F1-score, and 92.4% ROC-AUC, outperforming standalone CNN and Transformers. Segmentation was validated with ground truth, and SHAP highlighting modality-specific biomarkers. The model demonstrates strong generalizability across lesion types such as infarcts, white matter hyperintensities, and cerebral microbleeds supported by cross-validation and visualization. These findings underscore the model&s;s promise for real-world deployment in stroke and cerebrovascular disease.