Minding Fuzzy Regions: A Data-driven Alternating Learning Paradigm for Stable Lesion Segmentation
Lexin Fang, Yunyang Xu, Xiang Ma, Xuemei Li, Caiming Zhang
TL;DR
The paper tackles the challenge of segmenting fuzzy lesion regions with unclear boundaries and label noise in medical images. It introduces DALE, a data-driven alternating learning framework that first trains on reliable non-fuzzy regions to learn stable representations, then learns from fuzzy regions using a loss-consistency-based collaboration with adaptive label confidence and a distribution-alignment module. DALE combines an AE&ER soft-thresholding to partition data, a loss $\ ext{consistency}$-based optimization to adaptively weight labels via $\omega_t^{\star}$, and an unstable representation calibration guided by Wasserstein distance between class-conditional distributions. Empirical results across multiple datasets and backbones show consistent improvements in segmentation accuracy and stability, with notable gains in fuzzy-region performance, suggesting practical utility for robust lesion segmentation in clinical settings. The method is model-agnostic and comes with an open implementation, underscoring its potential to generalize to other medical-imaging segmentation tasks.
Abstract
Deep learning has achieved significant advancements in medical image segmentation, but existing models still face challenges in accurately segmenting lesion regions. The main reason is that some lesion regions in medical images have unclear boundaries, irregular shapes, and small tissue density differences, leading to label ambiguity. However, the existing model treats all data equally without taking quality differences into account in the training process, resulting in noisy labels negatively impacting model training and unstable feature representations. In this paper, a data-driven alternating learning (DALE) paradigm is proposed to optimize the model's training process, achieving stable and high-precision segmentation. The paradigm focuses on two key points: (1) reducing the impact of noisy labels, and (2) calibrating unstable representations. To mitigate the negative impact of noisy labels, a loss consistency-based collaborative optimization method is proposed, and its effectiveness is theoretically demonstrated. Specifically, the label confidence parameters are introduced to dynamically adjust the influence of labels of different confidence levels during model training, thus reducing the influence of noise labels. To calibrate the learning bias of unstable representations, a distribution alignment method is proposed. This method restores the underlying distribution of unstable representations, thereby enhancing the discriminative capability of fuzzy region representations. Extensive experiments on various benchmarks and model backbones demonstrate the superiority of the DALE paradigm, achieving an average performance improvement of up to 7.16%.