Non-iterative scribble-supervised learning with pacing pseudo-masks for medical image segmentation

Zefan Yang, Di Lin, Dong Ni, Yi Wang

Abstract: Highlights•A simple yet effective non-iterative method for scribble-supervised segmentation.•Entropy regularization to obtain high-confidence pseudo-masks for effective teaching.•Distorted augmentations to create discrepancy for consistency regularization.•A memory bank of ensemble features to complement scarce labeled supervision.•Competitive results compared with the fully-supervised counterparts.