Brain Tumor Segmentation Using a Multi-Route Convolutional Neural Network Based on an Attention Mechanism
The automatic segmentation of brain tumors in Magnetic Resonance Imaging (MRI) has been widely used and plays a key role in correct diagnosis and treatment planning. In this study, a four-path Convolutional Neural Network (CNN) is introduced that uses two attention modules in each path for segmenting brain tumors. The suggested attention module comprises a Channel Attention Block (CAB) to define more important channels among all input channels and a Spatial Attention Block (SAB) responsible for determining the most critical parts of an image including tumor areas. Moreover, a grouping of the Cross-entropy and Dice losses is used to address the problems of model convergence and data imbalance. Our technique was assessed on the BRATS2020 dataset, and the results indicate that the proposed model has competitive dice values; 0.863, 0.911, and 0.874 for the Edema, Enhanced, and Core tumor regions, respectively.
Keywords - Brain Tumor, Attention Mechanism, Channel Attention, Spatial Attention, Image Segmentation.