Exploring Smoothness and Class-Separation for Semi-supervised Medical Image Segmentation

Semi-supervised segmentation remains challenging in medical imaging since the amount of annotated data is often scarce and there are many blurred pixels near adhesive edges or low-contrast regions. To address issues, we advocate to firstly constrain consistency with without strong perturbations apply a sufficient smoothness constraint further encourage class-level separation exploit low-entropy regularization for model training. Particularly, this paper, propose SS-Net semi-supervised image tasks, via exploring pixel-level Smoothness inter-class Separation at same time. The forces generate invariant results under adversarial perturbations. Meanwhile, encourages individual class features should approach their corresponding high-quality prototypes, order make each distribution compact separate different classes. We evaluated our against five recent methods on public LA ACDC datasets. Extensive experimental two settings demonstrate superiority proposed model, achieving new state-of-the-art (SOTA) performance both code available https://github.com/ycwu1997/SS-Net .