Bridging the Inter-Domain Gap through Low-Level Features for Cross-Modal Medical Image Segmentation

2025-05-17

AI Summary

This paper addresses inter-domain discrepancy in unsupervised domain adaptation (UDA) for cross-modal medical image segmentation. To this end, we propose LowBridge—a lightweight, model-agnostic, plug-and-play two-stage style transfer framework that systematically exploits the cross-modal invariance of low-level geometric features (e.g., edges) shared between source and target domains. The method comprises three components: (i) decoupled source-domain training, (ii) edge-driven generative reconstruction on the target domain using a U-Net–GAN hybrid architecture, and (iii) edge-guided segmentation inference. Evaluated on multiple public benchmarks, LowBridge consistently outperforms 11 state-of-the-art UDA methods, achieving new SOTA performance. Ablation studies confirm its strong compatibility and generalizability across diverse generative backbones (e.g., U-Net, GAN variants) and segmentation architectures (e.g., nnU-Net, TransUNet).

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Segmentation
• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Systems and Infrastructure for Web, Mobile and WoT: Web applications in cross-disciplinary domains and verticals such as mixed reality, smart cities, and digital health
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

This paper addresses the task of cross-modal medical image segmentation by exploring unsupervised domain adaptation (UDA) approaches. We propose a model-agnostic UDA framework, LowBridge, which builds on a simple observation that cross-modal images share some similar low-level features (e.g., edges) as they are depicting the same structures. Specifically, we first train a generative model to recover the source images from their edge features, followed by training a segmentation model on the generated source images, separately. At test time, edge features from the target images are input to the pretrained generative model to generate source-style target domain images, which are then segmented using the pretrained segmentation network. Despite its simplicity, extensive experiments on various publicly available datasets demonstrate that proposed achieves state-of-the-art performance, outperforming eleven existing UDA approaches under different settings. Notably, further ablation studies show that proposed is agnostic to different types of generative and segmentation models, suggesting its potential to be seamlessly plugged with the most advanced models to achieve even more outstanding results in the future. The code is available at https://github.com/JoshuaLPF/LowBridge.

Problem

• Unsupervised domain adaptation for cross-modal medical image segmentation
• Leveraging shared low-level features to bridge inter-domain gaps
• Model-agnostic framework for generating and segmenting source-style target images

Innovation

• Uses edge features for cross-modal adaptation
• Generates source-style images via pretrained model
• Model-agnostic framework with modular components