CA-UNet: A Brain MRI Segmentation Model Based on U-Net with Attention Mechanism

Brain tissue segmentation is of paramount importance in the field of medical image processing, as its accuracy directly impacts subsequent diagnosis and treatment processes. However, the intricate structure of brain tissue makes it difficult to achieve precise segmentation. To address these issues, we propose a neural network based on UNet with attention mechanism and context fusion module, named CA-UNet. In the network, a Convolutional Block Attention Module (CBAM) is added in encoder and decoder to enhance the model's feature extraction capabilities. Then, the encoded information is sent to the Multi-scale Context Fusion Module (MCFM) for multi-scale context information fusion. In essence, CA-UNet refines the original UNet by incorporating attention mechanisms to better identify and segment small regions within brain MRI images, making it more effective for medical image analysis tasks. The results of this study indicate that the proposed method achieves an average Dice Similarity Coefficient of 95.77%, with white matter being 96.67%, cortical gray matter being 93.63%, basal ganglia and thalami being 95.00% and ventricular cerebrospinal fluid being 96.79%. The introduction of attention mechanism and contextual fusion module is an effective approach to enhance brain tissue segmentation performance.