TransMedSeg: A Transferable Semantic Framework for Semi-Supervised Medical Image Segmentation

2025-05-20

AI Summary

Cross-domain and cross-modality semantic knowledge transfer remains challenging in semi-supervised medical image segmentation due to domain shifts and modality heterogeneity. Method: We propose TransMedSeg, a transferable semantic framework centered on the novel Transferable Semantic Augmentation (TSA) module, which achieves implicit cross-domain semantic alignment without explicit data generation. We theoretically derive an upper bound on expected cross-entropy loss, enabling the first theory-driven semantic alignment optimization in semi-supervised learning (SSL). TransMedSeg integrates a teacher-student architecture, a lightweight memory module, cross-domain distribution matching, and intra-domain structural consistency preservation. Results: Evaluated on multi-center, multi-modality medical imaging benchmarks, TransMedSeg significantly outperforms state-of-the-art semi-supervised methods. It establishes a new paradigm for medical image representation learning that is both transferable across domains/modalities and label-efficient.

Technology Category

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

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Security and Privacy: Data transparency and provenance

Abstract

Semi-supervised learning (SSL) has achieved significant progress in medical image segmentation (SSMIS) through effective utilization of limited labeled data. While current SSL methods for medical images predominantly rely on consistency regularization and pseudo-labeling, they often overlook transferable semantic relationships across different clinical domains and imaging modalities. To address this, we propose TransMedSeg, a novel transferable semantic framework for semi-supervised medical image segmentation. Our approach introduces a Transferable Semantic Augmentation (TSA) module, which implicitly enhances feature representations by aligning domain-invariant semantics through cross-domain distribution matching and intra-domain structural preservation. Specifically, TransMedSeg constructs a unified feature space where teacher network features are adaptively augmented towards student network semantics via a lightweight memory module, enabling implicit semantic transformation without explicit data generation. Interestingly, this augmentation is implicitly realized through an expected transferable cross-entropy loss computed over the augmented teacher distribution. An upper bound of the expected loss is theoretically derived and minimized during training, incurring negligible computational overhead. Extensive experiments on medical image datasets demonstrate that TransMedSeg outperforms existing semi-supervised methods, establishing a new direction for transferable representation learning in medical image analysis.

Problem

• Enhancing semi-supervised medical image segmentation with transferable semantics
• Addressing overlooked cross-domain semantic relationships in SSL methods
• Improving feature representation via domain-invariant alignment and structural preservation

Innovation

• Transferable Semantic Augmentation module enhances features
• Aligns domain-invariant semantics via cross-domain matching
• Implicit semantic transformation with lightweight memory module