ShapePU: A New PU Learning Framework Regularized by Global Consistency for Scribble Supervised Cardiac Segmentation

TL;DR

ShapePU introduces a novel weakly supervised cardiac segmentation method using scribble annotations, PU learning, and global consistency regularization, achieving superior performance over fully supervised and existing weakly supervised methods.

Contribution

The paper proposes ShapePU, a new PU learning framework with global consistency regularization for scribble-supervised cardiac segmentation, effectively leveraging unlabeled pixels and shape knowledge.

Findings

Outperforms fully supervised methods by 1.4% and 9.8% in Dice score on two datasets.

Surpasses state-of-the-art weakly supervised and PU learning methods significantly.

Demonstrates effectiveness of global consistency and shape regularization in weak supervision.

Abstract

Cardiac segmentation is an essential step for the diagnosis of cardiovascular diseases. However, pixel-wise dense labeling is both costly and time-consuming. Scribble, as a form of sparse annotation, is more accessible than full annotations. However, it's particularly challenging to train a segmentation network with weak supervision from scribbles. To tackle this problem, we propose a new scribble-guided method for cardiac segmentation, based on the Positive-Unlabeled (PU) learning framework and global consistency regularization, and termed as ShapePU. To leverage unlabeled pixels via PU learning, we first present an Expectation-Maximization (EM) algorithm to estimate the proportion of each class in the unlabeled pixels. Given the estimated ratios, we then introduce the marginal probability maximization to identify the classes of unlabeled pixels. To exploit shape knowledge, we apply cutout operations to training images, and penalize the inconsistent segmentation results. Evaluated on two open datasets, i.e, ACDC and MSCMRseg, our scribble-supervised ShapePU surpassed the fully supervised approach respectively by 1.4% and 9.8% in average Dice, and outperformed the state-of-the-art weakly supervised and PU learning methods by large margins. Our code is available at https://github.com/BWGZK/ShapePU.