An Evidential-enhanced Tri-Branch Consistency Learning Method for Semi-supervised Medical Image Segmentation

TL;DR

ETC-Net tackles the challenges of semi-supervised medical image segmentation by integrating evidential learning into a tri-branch consistency framework. It introduces an Evidential Conservative Branch (ECB) and an Evidential Progressive Branch (EPB) to preserve prediction disagreement and reliability, and an Evidential Fusion Branch (EFB) to fuse their outputs via Dempster-Shafer theory. Uncertainty-guided bidirectional cross supervision (BUCS) and evidence-based fusion yield high-quality pseudo-labels for unlabeled data, improving segmentation accuracy on LA, Pancreas-CT, and ACDC with limited annotations. The approach achieves state-of-the-art performance across multiple metrics and datasets, demonstrating strong practical potential for clinical-grade semi-supervised segmentation in resource-constrained labeling scenarios. The method emphasizes uncertainty-aware learning and evidence fusion to enhance robustness and generalization in medical image analysis.

Abstract

Semi-supervised segmentation presents a promising approach for large-scale medical image analysis, effectively reducing annotation burdens while achieving comparable performance. This methodology holds substantial potential for streamlining the segmentation process and enhancing its feasibility within clinical settings for translational investigations. While cross-supervised training, based on distinct co-training sub-networks, has become a prevalent paradigm for this task, addressing critical issues such as predication disagreement and label-noise suppression requires further attention and progress in cross-supervised training. In this paper, we introduce an Evidential Tri-Branch Consistency learning framework (ETC-Net) for semi-supervised medical image segmentation. ETC-Net employs three branches: an evidential conservative branch, an evidential progressive branch, and an evidential fusion branch. The first two branches exhibit complementary characteristics, allowing them to address prediction diversity and enhance training stability. We also integrate uncertainty estimation from the evidential learning into cross-supervised training, mitigating the negative impact of erroneous supervision signals. Additionally, the evidential fusion branch capitalizes on the complementary attributes of the first two branches and leverages an evidence-based Dempster-Shafer fusion strategy, supervised by more reliable and accurate pseudo-labels of unlabeled data. Extensive experiments conducted on LA, Pancreas-CT, and ACDC datasets demonstrate that ETC-Net surpasses other state-of-the-art methods for semi-supervised segmentation. The code will be made available in the near future at https://github.com/Medsemiseg.

Figures (9)

• Figure 1: (a) The traditional cross-supervised learning way. (b) Our proposed ETC-Net. The traditional method is prone to ineffective cross-supervised training when there is no difference between two predictions. Our method can alleviate this issue by generating complementary and reliable cross-supervision. Note that the white color in predictions represents ground truth.
• Figure 2: The flowchart of the proposed ETC-Net. $\otimes$ means pixel-wise weighting by uncertain maps $u_i$. $\oplus$ denotes the fusion operation.
• Figure 3: The calculation process of ETC-Net. The orange arrows represent the direction of cross-supervised training.
• Figure 4: 2D and 3D Visualized results of semi-supervised segmentation results on LA dataset for left atrium segmentation.
• Figure 5: 2D and 3D Visualized results of semi-supervised segmentation results for pancreas segmentation.