Uncertainty-Guided Dual-Domain Learning for Reliable Skin Lesion Segmentation

2026-05-10

AI Summary

This work addresses the challenges of spatial modeling and overfitting in skin lesion segmentation caused by visual ambiguity, irregular morphology, and label noise. To this end, we propose UGDD-Net, which pioneers the use of pixel-wise prediction uncertainty as an active guidance signal in both spatial and spectral domains. The framework incorporates uncertainty-guided bidirectional feature fusion (UGBFF), uncertainty-guided graph refinement (UGGR), and an uncertainty-guided boundary-adaptive loss (UGML) to enable topology-aware feature interaction and robust optimization. Extensive experiments on ISIC2017, ISIC2018, PH2, and HAM10000 demonstrate state-of-the-art performance, with particularly strong results on challenging cases. Moreover, the generated uncertainty maps exhibit high alignment with discrepancies in expert annotations, significantly enhancing model robustness and interpretability.

Abstract

Accurate skin lesion segmentation is vital for dermoscopic Computer-Aided Diagnosis. However, visual ambiguity and morphological irregularity often defeat spatial modeling, necessitating multi-domain architectures. Existing paradigms frequently overlook the active use of prediction uncertainty, leading to deterministic frameworks that suffer from blind cross-domain fusion and overfit to label noise. To address these issues, we propose the Uncertainty-Guided Dual-Domain Network (UGDD-Net). UGDD-Net introduces a novel "Glance-and-Gaze" mechanism to transform uncertainty into an active guiding signal. Specifically, the Uncertainty-Guided Bi-directional Feature Fusion (UGBFF) module uses pixel-level uncertainty to modulate spatial-spectral interactions. The Uncertainty-Guided Graph Refinement (UGGR) module constructs a topology-aware graph to propagate reliable semantic consensus and refine uncertain nodes. Finally, the Uncertainty-Guided Margin-Adaptive Loss (UGML) enforces strict constraints on confident pixels while relaxing penalties on uncertain ones to improve statistical calibration. Extensive experiments on ISIC2017, ISIC2018, PH2, and HAM10000 datasets demonstrate that UGDD-Net achieves state-of-the-art performance, especially on "Hard Samples". Our uncertainty maps align with expert inter-observer variability, providing robust interpretability for human-machine collaborative diagnosis.

Problem

• skin lesion segmentation
• prediction uncertainty
• multi-domain learning
• label noise
• cross-domain fusion

Innovation

• Uncertainty-Guided Learning
• Dual-Domain Fusion
• Graph-Based Refinement
• Adaptive Loss Function
• Skin Lesion Segmentation