Deep convolutional level set network for medical image segmentation

A deep convolution level set network for medical image segmentation constructs a new parallel cavity convolution sequence module in a pre-segmentation part of a U-Net model, captures local and global feature information at the same time, enhances feature extraction, and minimizes loss of image information in a downsampling process; a mixed attention module is constructed at a decoder end, the most important features can be automatically learned, selected and weighted, the influence of redundant information is reduced, and network upsampling is helped to recover information and reconstruct images; taking the output of the U-Net as a traction item of an energy function in a level set method, taking the level set method as a post-processing step, and receiving prior information from the U-Net; a distance regular term and an edge stopping function based on a logarithmic function and a polynomial function are constructed, and a segmentation image boundary is further optimized by predicting edge information of a ta