Brain Tumor Detection and Segmentation Using Deep Learning Models with Dual Attention Mechanism

Brain tumor segmentation is a critical task in medical image analysis with applications in diagnosis, treatment planning, and patient monitoring. Deep learning models have shown remarkable success in automating this process. This research presents an investigation into the integration of dual attention mechanisms into prominent deep learning architectures, namely TransFuse-L, U-Net, and ResNetU++. The selected deep architectures are evaluated with and without dual attention mechanism. They are trained and tested on BRaTS-2021 Task-1 Dataset which is a benchmark for brain tumor segmentation. Dual attention mechanisms, encompassing both spatial and channel attention, offer the potential to simultaneously capture local details and global context within medical images. The motivation behind this research lies in enhancing the interpretability and performance of existing models by enabling them to focus on salient features while considering broader image context. The experimental results demonstrate that the integration of dual attention mechanisms consistently leads to improved brain tumor segmentation performance across the selected deep architectures. The findings underscore the utility of dual attention mechanisms in enhancing the interpretability and predictive capabilities of deep learning models for brain tumor segmentation.