A Hybrid UNet Model Incorporating Attention Mechanism for Adaptive Brain Tumor Magnetic Resonance Segmentation

The segmentation of medical images is crucial, particularly in brain tumor MR imaging, as it aids doctors in accurate diagnosis and treatment planning. However, conventional UNet models often face limitations due to their inability to capture intricate features, resulting in suboptimal segmentation performance. This study addresses these challenges by proposing a UNet model enhanced with a Convolutional Block Attention Module (CBAM). This advanced network leverages spatial and channel attention mechanisms that work together to focus more effectively on important image regions, refining feature representation. Experiments conducted on the TCGA brain tumor MR dataset demonstrate that our proposed UNet+CBAM model outperforms the standard UNet, achieving a 2.5% improvement in the Dice coefficient. This significant enhancement underscores the efficacy of integrating attention mechanisms into medical image segmentation. We conclude that our UNet+CBAM model offers an improved and reliable solution for brain tumor segmentation, potentially advancing its clinical application.