Risk-Routed Implicit Boundary Refinement for Robust Ultrasound Image Segmentation
arXiv:2607.21787 (cs.CV)
Submitted on 23 Jul 2026
Authors: Jingguo Qu, Xinyang Han, Xiang Wang, Yuqi Yang, Tonghuan Xiao, Sheng Ning, Jing Qin, Ann Dorothy King, Winnie Chiu-Wing Chu, Jing Cai, Michael Ying
Abstract
Medical ultrasound (US) image segmentation remains challenging due to speckle noise, low-contrast boundaries, acoustic shadowing, and variability in acquisition across operators and clinical centers. While encoder-decoder and transformer-based networks have achieved strong performance, many existing methods recover boundary details through dense decoders or larger backbones, often producing over-smoothed contours or unstable predictions under external distribution shifts.
In this paper, we propose Risk-routed Implicit Boundary Refinement (RIBR), a compact segmentation framework that leverages implicit neural representation as a risk-routed residual correction rather than an unconstrained full-mask predictor. RIBR integrates three key components:
- Boundary-refinement implicit residuals that enhance contour accuracy
- Risk-routed residual control to selectively apply corrections only where needed
- Geometry- and speckle-aware boundary regularization to suppress non-boundary oscillations
We evaluated RIBR on nine ultrasound datasets covering lymph nodes, breast lesions, thyroid nodules, and prostate. The method achieves the best overall macro-average performance and consistently reduces boundary error across grouped and organ-specific comparisons—all while maintaining a compact parameter budget.
Our findings demonstrate that controlled implicit residual learning is a practical and effective strategy for resource-constrained and boundary-sensitive ultrasound segmentation. Source code is available at https://github.com/jinggqu/ribr.
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2607.21787 [cs.CV]
Published as part of the arXiv preprint series (July 2026). This version reflects the latest advances in robust medical image segmentation using implicit neural representations.
via ArXiv CV