Semantic Segmentation of Brain MRI Based on U-net Network and Edge Loss

Brain MRI analysis is of great significance for extracting clinical information from patients and providing diagnostic recommendations for doctors. However, brain MRI is difficult to detect and segment because of its complex structure and great difference. To deal with these problems, in the field of medical image semantic segmentation, the model based on U-net structure in the depth neural network model has excellent performance. In this work, in order to increased the influence of image edge pixels on segmentation results and improved the accuracy of medical image segmentation, we provided a more optimized U-net model that can be applied to medical image semantic segmentation. The model integrated the convolution block attention module and after the feature extraction, combined with the edge detection network, the segmentation effect of edge pixels was enhanced. At the same time, by improving the residual convolution block, the amount of parameters was greatly reduced. On the BraTS-2017 data set, we had good experiment results.