Semantic Class Distribution Learning for Debiasing Semi-Supervised Medical Image Segmentation

2026-03-05

AI Summary

This work addresses the challenge of severe class imbalance in semi-supervised medical image segmentation, where minority anatomical structures are often overwhelmed by dominant classes. To mitigate this issue, the authors propose a Semantic Class Distribution Learning (SCDL) framework that explicitly models class-conditional feature distributions. The approach introduces a plug-and-play SCDL module incorporating Class Distribution Bidirectional Alignment (CDBA) and Semantic Anchor Constraints (SAC) to effectively decouple and optimize the feature distributions of individual classes, thereby alleviating both supervision bias and representation bias. Experimental results on the Synapse and AMOS datasets demonstrate that the proposed method significantly improves overall and class-wise segmentation performance, with particularly notable gains for underrepresented minority classes.

Technology Category

• Machine Learning: Semi-Supervised Learning
• Computer Vision: Segmentation
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

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: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Medical image segmentation is critical for computer-aided diagnosis. However, dense pixel-level annotation is time-consuming and expensive, and medical datasets often exhibit severe class imbalance. Such imbalance causes minority structures to be overwhelmed by dominant classes in feature representations, hindering the learning of discriminative features and making reliable segmentation particularly challenging. To address this, we propose the Semantic Class Distribution Learning (SCDL) framework, a plug-and-play module that mitigates supervision and representation biases by learning structured class-conditional feature distributions. SCDL integrates Class Distribution Bidirectional Alignment (CDBA) to align embeddings with learnable class proxies and leverages Semantic Anchor Constraints (SAC) to guide proxies using labeled data. Experiments on the Synapse and AMOS datasets demonstrate that SCDL significantly improves segmentation performance across both overall and class-level metrics, with particularly strong gains on minority classes, achieving state-of-the-art results. Our code is released at https://github.com/Zyh55555/SCDL.

Problem

• class imbalance
• semi-supervised learning
• medical image segmentation
• minority classes
• feature representation bias

Innovation

• Semantic Class Distribution Learning
• Class Distribution Bidirectional Alignment
• Semantic Anchor Constraints
• Debiasing
• Semi-Supervised Medical Image Segmentation