Collaborative boundary-aware context encoding networks for error map prediction

Accurately assessing the medical image segmentation quality of automatically generated predictions is essential for guaranteeing reliability results computer-assisted diagnosis (CAD). Many researchers have studied estimation without labeled ground truths. Recently, a novel idea proposed, which transforms assessment (SQA) into pixel-wise or voxel-wise error map task. However, simple application vanilla structures in domain fails to achieve satisfactory results. In this paper, we propose collaborative boundary-aware context encoding networks called EP-Net Specifically, feature transformation branch better fusion between images and masks, precise localization regions. Further, module utilize global predictor from enhance representation regularize networks. Extensive experiments on IBSR V2.0 dataset, ACDC dataset M&Ms demonstrate that achieves compared with traditional patterns. Based prediction results, obtain proxy metric quality, has high Pearson correlation coefficient real accuracy all datasets.