Semisupervised Medical Image Segmentation through Prototype-Based Mutual Consistency Learning

Figure 1

Overview of the proposed prototype-based mutual consistency learning network (PMCL) for semisupervised medical image segmentation. The entire framework consists of two branches, inspired by the Unet [13] method. The orange line represents the flow of labeled images, and the blue line represents the flow of unlabeled images. The entire framework can be divided into three parts. (a) Embed depth features into the original image. (b) Perform segmentation tasks on student models. (c) The composition of the framework loss we proposed.