Implicit Fuzzification via Bounded Noise Injection for Robust Medical Image Segmentation

2026-06-03

AI Summary

This work addresses the challenge of ambiguous boundaries in medical image segmentation, which often arises from sampling artifacts and annotation uncertainty, while existing models tend to be overconfident and neglect the inherent fuzziness of transition regions. To tackle this issue, we propose NoiseUNet, a novel architecture that injects bounded noise perturbations into the skip connections of a U-Net to implicitly induce fuzziness, thereby enabling a data-driven soft membership representation without explicit modeling. This approach enhances the robustness of multi-scale feature fusion and improves boundary awareness. We introduce ThyR, the first thyroid ultrasound dataset with real-world fuzzy boundaries, and demonstrate through comprehensive experiments that our method achieves significant gains in both segmentation accuracy and boundary fidelity, establishing its effectiveness and superiority in handling ambiguous boundaries in medical images.

Abstract

Image segmentation remains fundamentally limited by boundary ambiguity arising from sampling-induced information loss and inherent uncertainty in pixel-wise labeling. Although encoder-decoder architectures such as U-Net achieve strong performance, they often produce overconfident predictions that fail to capture transition-region ambiguity. To address this issue, we propose \textbf{NoiseUNet}, a simple yet effective framework that injects bounded perturbations into skip connections to regularize cross-scale feature fusion. This mechanism enforces robustness to local feature variations and promotes boundary-aware representations. Theoretically, the perturbation induces an implicit fuzzification effect, yielding soft, data-driven memberships without requiring explicit fuzzy modeling. We further introduce \textbf{ThyR}, a real-world thyroid ultrasound dataset with inherently ambiguous boundaries. Experiments demonstrate that NoiseUNet consistently improves both segmentation accuracy and boundary fidelity.

Problem

• boundary ambiguity
• medical image segmentation
• uncertainty
• fuzzification
• pixel-wise labeling

Innovation

• implicit fuzzification
• bounded noise injection
• boundary-aware segmentation
• skip connection perturbation
• medical image segmentation