NMDAU-Net: A Novel Lightweight 3D Network for Precision Segmentation of Brain Gliomas in MRI

Brain MRI images are inherently three-dimensional, and traditional segmentation methods frequently fail to capture critical information. To address the complexities of 3D brain glioma MRI image segmentation, we introduced NMDAU-Net, a high-performance lightweight 3D segmentation network. This network builds upon the 3D U-Net architecture by integrating an enhanced 3D decomposable convolution block and dense attention modules (DAMs), significantly improving feature interaction and representation. Incorporating the avoid space pyramid pooling (ASPP) module as a transition structure between the encoder and decoder further augments feature extraction and enables the capture of richer semantic information. In addition, a weighted bidirectional feature pyramid module replaces the conventional skip connections in the 3D U-Net, facilitating the integration of multiscale features. Our model was evaluated on a dataset comprising more than 378 3D brain glioma MRI images and achieved a Dice score of 86.91%. The enhanced segmentation precision of NMDAU-Net offers crucial support for precise diagnosis and personalized treatment strategies and is promising for significantly improving treatment outcomes for glioma. This demonstrates its substantial potential for clinical application in enhancing patient prognosis and survival rates.