U-shaped brain tumor segmentation network fused with cavity convolution

In order to solve the problem that a classic U-shaped segmentation network is not high in tumor sub-region segmentation boundary precision, the invention provides a U-shaped network (Dilated Convolution Unet, DCU-Net) fused with cavity convolution. On the basis of a classic U-Net structure, the algorithm comprises: firstly, cutting and preprocessing an MR brain tumor image, and relieving the classimbalance problem by reducing input of background pixels; replacing maximum pooling at the tail end of the contraction path with multi-scale spatial pyramid pooling, and expanding the feature receptive field while maintaining the resolution; and finally, introducing a hole convolution residual block through Add operation to improve jump connection in the training network, and fusing low-level features from a contraction path. The ability of the network to identify tumor details is improved, so that a more accurate brain tumor segmentation result is obtained. The network has a good applicationprospect in automatic seg