The Segmentation of Brain Tumor from MR Images Based on U-Net++ Network and Attention Mechanism

Glioma is a malignant brain tumor that seriously endangers people's lives. Automatic segmentation of brain tumor from magnetic resonance (MR) images plays an important role in accurately diagnosing this disease and formulating corresponding treatment plans. Due to the problems of blurred field of view and unclear image edges in MR images of brain tumors, segmentation using separate U-Net or U-Net++ networks often does not yield accurate results. In this paper, we propose a segmentation model that effectively combines residual module, multiscale deep supervision, convolutional block attention module (CBAM) and U-Net++, named improved CBAM-U-Net++. It can capture features at different levels and dimensions of the feature image, and at the same time circumvents the problem of the encoder degradation during the convolution process. The use of multimodal MR image data can overcome the detection limitations of unimodal, fully take into account the relevant information of the lesion region, and improve the accuracy of segmentation. Experimental results on the BraTS2020 dataset show that the proposed model can further improve the accuracy of brain tumor segmentation compared to U-Net or U-Net++ networks.