Volumetric Densely Dilated Spatial Pooling ConNets for Prostate Segmentation

The high incidence rate of prostate disease poses a need in early detection for diagnosis. As one of the main imaging methods used for prostate cancer detection, magnetic resonance imaging (MRI) has wide range in appearance and imbalance problems, making automated prostate segmentation fundamental but challenging we here propose a novel Densely Dilated Spatial Pooling ConNets (DDSP) in encoder-decoder structure. It employs dense structure to combine dilated convolution and global pooling, thus supplies the coarse segmentation results produced by encoder and decoder subnet and preserves more contextual information. To obtain hierarchical feature maps, residual long connection is furtherly adopted to fuse contexture features. Meanwhile, we adopt DSC loss and Jaccard loss functions to train our DDSP network. We surprisingly found and proved that, in contrast to (re-weighted) cross entropy, DSC loss and Jaccard loss have a lot of benign properties in theory, including symmetry, continuity and differentiation about the parameters of network. Extensive experiments on the MICCAI PROMISE12 challenge dataset have been done to corroborate the effectiveness of our DDSP network with DSC loss and Jaccard loss; totally, in test dataset our method achieves a score of 85.78, outperforming most of other competitors.