A mutual communicated model based on multi-parametric MRI for automated prostate segmentation and prostate cancer classification

Deep learning methods for multi-parametric MRI hold the greatest promise for automated computer-aided diagnosis of prostate cancer, including classification and segmentation. In this work, we propose a new model (MC-DSCN) for classification and segmentation simultaneously. MC-DSCN contains three components: the coarse segmentation component based on the residual U-net with attention blocks, the classification component based on the stacked residual blocks and multi-parametric fusion mechanism, and the fine segmentation component that incorporates the information about lesion location (cancer response map, CRM) arising from the classification component. Extensive experiments are performed to demonstrate that the proposed method could effectively transfer mutual information between segmentation and classification components and facilitate each other in a bootstrapping way, thus outperforming the methods designed to perform only one task.