Self-supervised Correction Learning for Semi-supervised Biomedical Image Segmentation

Biomedical image segmentation plays a significant role in computer-aided diagnosis. However, existing CNN based methods rely heavily on massive manual annotations, which are very expensive and require huge human resources. In this work, we adopt coarse-to-fine strategy propose self-supervised correction learning paradigm for semi-supervised biomedical segmentation. Specifically, design dual-task network, including shared encoder two independent decoders lesion region inpainting, respectively. the first phase, only branch is used to obtain relatively rough result. second step, mask detected regions original initial map, send it together with into network again simultaneously perform inpainting separately. For labeled data, process supervised by unlabeled guided loss of masked regions. Since tasks similar feature information, data effectively enhances representation further improves performance. Moreover, gated fusion (GFF) module designed incorporate complementary features from tasks. Experiments three medical datasets different polyp, skin fundus optic disc well demonstrate outstanding performance our method compared other approaches. The code available at https://github.com/ReaFly/SemiMedSeg.