A novel deep learning model DDU-net using edge features to enhance brain tumor segmentation on MR images

Highlights：

• We propose DDU-net to improve the performance by making the network focus on the learning of edge features.

• We use a novel edge extraction algorithm to ensure each pixel on one edge comes from the same part of the tumor.

• We propose an improved CE loss function to solve the serious class imbalance problem in the learning of edges.

• We introduce a regularization loss to encourage the predicted segmentation map to align with the boundary of ground truth.