Soft-Hard Attention Network for Pseudo-Color Mri Image Segmentation

Deep learning has achieved excellent results in the tumor image segmentation, and the application of the attention mechanism has made the model more powerful. In this paper, a novel Soft-Hard Method (SHM) is proposed, which can transform the current soft attention and possess some characteristics of hard attention. Then, we propose a Soft-Hard Attention Network (SHANet), which integrates the Soft-Hard attention we modified. Finally, we verify the effectiveness of SHM and the performance of SHANet in the low-grade glioma Pseudo-color MRI Image Segmentation task.