MSDC-Net: a corpus callosum MRI segmentation network based on multi-scale and dense connection technologies

As the largest white matter structure in the human brain, the corpus callosum is composed of nerve fibers connecting the left and right hemispheres, serving as a key pathway for information transmission between the two hemispheres. Its morphological changes are closely related to neurodegenerative diseases, inflammatory diseases, and other conditions. Accurate automatic segmentation of the corpus callosum is of great significance for assisting clinical diagnosis and promoting the development of intelligent diagnostic technologies in medical imaging. To address issues such as uneven pixel value distribution, blurred boundaries, and individual differences in corpus callosum MRI images, this paper proposes a segmentation method based on multi-scale and dense connection technologies. The main contributions include: (1) Designing a Deep Stacking Module (DSM), which enhances the feature extraction capability in the encoding stage by improving the feature fusion method of dense connections, thereby enhancing the network's noise robustness and reducing runtime; (2) Constructing a Multi-channel Spatial Attention Module (MSAM), which is embedded into skip connections to strengthen the recovery of low-level positional features and the selection of contextual information, while reducing feature redundancy. Experiments on the OASIS-1 show that the three-channel multi-scale spatial attention module achieves the optimal segmentation performance, with the DICE index of the proposed network reaching 94.33%. The research results confirm that this method can accurately extract the corpus callosum region, providing an efficient tool for the automated diagnosis and treatment of brain diseases.