Mutual Evidential Deep Learning for Medical Image Segmentation

TL;DR

This paper introduces MEDL, a novel semi-supervised medical image segmentation framework that leverages multiple networks and evidential fusion to improve pseudo-label reliability and model performance.

Contribution

MEDL combines diverse architecture networks with evidential fusion and uncertainty-based learning to enhance pseudo-label quality in semi-supervised segmentation.

Findings

Achieves state-of-the-art results on five datasets.

Effectively handles pseudo-label uncertainty and bias.

Improves segmentation accuracy with evidential fusion.

Abstract

Existing semi-supervised medical segmentation co-learning frameworks have realized that model performance can be diminished by the biases in model recognition caused by low-quality pseudo-labels. Due to the averaging nature of their pseudo-label integration strategy, they fail to explore the reliability of pseudo-labels from different sources. In this paper, we propose a mutual evidential deep learning (MEDL) framework that offers a potentially viable solution for pseudo-label generation in semi-supervised learning from two perspectives. First, we introduce networks with different architectures to generate complementary evidence for unlabeled samples and adopt an improved class-aware evidential fusion to guide the confident synthesis of evidential predictions sourced from diverse architectural networks. Second, utilizing the uncertainty in the fused evidence, we design an asymptotic Fisher information-based evidential learning strategy. This strategy enables the model to initially focus on unlabeled samples with more reliable pseudo-labels, gradually shifting attention to samples with lower-quality pseudo-labels while avoiding over-penalization of mislabeled classes in high data uncertainty samples. Additionally, for labeled data, we continue to adopt an uncertainty-driven asymptotic learning strategy, gradually guiding the model to focus on challenging voxels. Extensive experiments on five mainstream datasets have demonstrated that MEDL achieves state-of-the-art performance.