Learning Semantic Directions for Feature Augmentation in Domain-Generalized Medical Segmentation

2025-07-31

AI Summary

Medical image segmentation suffers from domain shift across clinical centers due to variations in imaging equipment and acquisition protocols, severely degrading model generalizability. To address this challenge, we propose a generic domain feature enhancement framework. First, a learnable semantic direction selector is designed to emphasize anatomy-consistent features. Second, a covariance-driven intensity sampling mechanism dynamically models and modulates domain-variant components. Third, implicit feature perturbation coupled with adaptive consistency constraints stabilizes cross-domain predictions. Crucially, our method operates without access to target-domain data. Evaluated on two multi-center public benchmarks, it outperforms state-of-the-art domain generalization methods, achieving average improvements of 3.2–5.7 percentage points in the Sørensen–Dice coefficient. These results demonstrate both the efficacy and practical applicability of our approach for robust segmentation in unseen clinical settings.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Application Domains: Security

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Medical image segmentation plays a crucial role in clinical workflows, but domain shift often leads to performance degradation when models are applied to unseen clinical domains. This challenge arises due to variations in imaging conditions, scanner types, and acquisition protocols, limiting the practical deployment of segmentation models. Unlike natural images, medical images typically exhibit consistent anatomical structures across patients, with domain-specific variations mainly caused by imaging conditions. This unique characteristic makes medical image segmentation particularly challenging. To address this challenge, we propose a domain generalization framework tailored for medical image segmentation. Our approach improves robustness to domain-specific variations by introducing implicit feature perturbations guided by domain statistics. Specifically, we employ a learnable semantic direction selector and a covariance-based semantic intensity sampler to modulate domain-variant features while preserving task-relevant anatomical consistency. Furthermore, we design an adaptive consistency constraint that is selectively applied only when feature adjustment leads to degraded segmentation performance. This constraint encourages the adjusted features to align with the original predictions, thereby stabilizing feature selection and improving the reliability of the segmentation. Extensive experiments on two public multi-center benchmarks show that our framework consistently outperforms existing domain generalization approaches, achieving robust and generalizable segmentation performance across diverse clinical domains.

Problem

• Addressing domain shift in medical image segmentation
• Enhancing robustness to imaging condition variations
• Improving generalizability across diverse clinical domains

Innovation

• Implicit feature perturbations guided by domain statistics
• Learnable semantic direction selector for feature modulation
• Adaptive consistency constraint stabilizing feature selection