SP${ }^3$ : Superpixel-propagated pseudo-label learning for weakly semi-supervised medical image segmentation

TL;DR

A SuperPixel-Propagated Pseudo-label learning method, using the structural information contained in superpixel for supplemental information for weakly semi-supervised segmentation and aiming to alleviate the negative impact of noise in pseudo-label, superpixel-level uncertainty is incorporated to guide the pseudo-label supervision for stable learning.

Abstract

Deep learning-based medical image segmentation helps assist diagnosis and accelerate the treatment process while the model training usually requires large-scale dense annotation datasets. Weakly semi-supervised medical image segmentation is an essential application because it only requires a small amount of scribbles and a large number of unlabeled data to train the model, which greatly reduces the clinician's effort to fully annotate images. To handle the inadequate supervisory information challenge in weakly semi-supervised segmentation (WSSS), a SuperPixel-Propagated Pseudo-label (SP${}^3$) learning method is proposed, using the structural information contained in superpixel for supplemental information. Specifically, the annotation of scribbles is propagated to superpixels and thus obtains a dense annotation for supervised training. Since the quality of pseudo-labels is limited by the low-quality annotation, the beneficial superpixels selected by dynamic thresholding are used to refine pseudo-labels. Furthermore, aiming to alleviate the negative impact of noise in pseudo-label, superpixel-level uncertainty is incorporated to guide the pseudo-label supervision for stable learning. Our method achieves state-of-the-art performance on both tumor and organ segmentation datasets under the WSSS setting, using only 3\% of the annotation workload compared to fully supervised methods and attaining approximately 80\% Dice score. Additionally, our method outperforms eight weakly and semi-supervised methods under both weakly supervised and semi-supervised settings. Results of extensive experiments validate the effectiveness and annotation efficiency of our weakly semi-supervised segmentation, which can assist clinicians in achieving automated segmentation for organs or tumors quickly and ultimately benefit patients.

Figures (7)

• Figure 1: a) Sample images of scribbles, ground truth, and superpixels. b) Illustration of superpixel quality. c) Illustration of prediction quality.
• Figure 2: a) An overview of the proposed method. b) Illustration of superpixel-based scribble expansion. c) Illustration of pseudo-label refinement with superpixel filtered by dynamic threshold. d) Illustration of superpixel uncertainty assessment.
• Figure 3: Visual comparison of weakly semi-supervised segmentation results for the ACDC datasets under different labeled ratios.
• Figure 4: Visual comparison of weakly semi-supervised segmentation results for the BraTS2019 datasets under different labeled ratios.
• Figure 5: Quantitative analysis in iteration. We present the training curves on fold1 of the ACDC dataset with a 10% annotation ratio as follows: a) Dice performance of ablation experiments. LB represents the lower bound, E corresponds to E_scri, P corresponds to Pseu, U corresponds to Unce. b) Dice performance of comparative methods. c) Dynamic threshold of different class. d) Class-average superpixel sampling rate, representing the proportion of superpixels selected based on the dynamic threshold.