Dynamic Focus on Tumor Boundaries: A Lightweight U-Net for MRI Brain Tumor Segmentation

Accurate segmentation of brain tumors in MRI scans is critical for diagnosis and treatment planning. Traditional segmentation models, such as U-Net, excel in capturing spatial information but often struggle with complex tumor boundaries and subtle variations in image contrast. These limitations can lead to inconsistencies in identifying critical regions, impacting the accuracy of clinical outcomes. To address these challenges, this paper proposes a novel modification to the U-Net architecture by integrating a spatial attention mechanism designed to dynamically focus on relevant regions within MRI scans. This innovation enhances the model's ability to delineate fine tumor boundaries and improves segmentation precision. Our model was evaluated on the Figshare dataset, which includes annotated MRI images of meningioma, glioma, and pituitary tumors. The proposed model achieved a Dice similarity coefficient (DSC) of 0.93, a recall of 0.95, and an AUC of 0.94, outperforming existing approaches such as V-Net, DeepLab V3+, and nnU-Net. These results demonstrate the effectiveness of our model in addressing key challenges like low-contrast boundaries, small tumor regions, and overlapping tumors. Furthermore, the lightweight design of the model ensures its suitability for real-time clinical applications, making it a robust tool for automated tumor segmentation. This study underscores the potential of spatial attention mechanisms to significantly enhance medical imaging models and paves the way for more effective diagnostic tools.