Beyond Random Sampling: Distribution-Aware Alignment for Semi-Supervised Medical Image Segmentation

2026-07-05

AI Summary

This work addresses the challenge of semi-supervised medical image segmentation under extremely low annotation rates, where random sampling often overlooks data distribution heterogeneity, leading to biased representations. To mitigate this, the authors propose an efficient distribution-aligned framework that first constructs a more representative labeled set via offline distribution-aware sampling based on the Density-K-Center algorithm. They further introduce a memory-guided copy-paste module and a progressive easy-to-hard pseudo-labeling strategy to enhance cross-domain semantic consistency and suppress noise. By moving beyond the conventional i.i.d. assumption and integrating vision foundation models with a semantic memory mechanism, the method achieves state-of-the-art performance across six 2D and 3D medical imaging datasets, demonstrating particularly strong results in extreme low-resource settings such as a 1/16 annotation rate.

Abstract

Precise medical image segmentation is crucial for clinical diagnosis and treatment planning, yet relies heavily on expensive expert annotations. Semi-supervised medical image segmentation (SSMIS) offers a cost-effective solution but typically operates under the assumption of independent and identically distributed (i.i.d.) data, defaulting to random sampling. While statistically valid at scale, this strategy suffers from severe representation bias in low-data regimes, failing to capture the heterogeneous medical data manifold. To address this, we propose a highly data-efficient framework driven by distribution alignment. First, we introduce an offline Distribution-Aware Sample Selection strategy. By leveraging Vision Foundation Models (VFMs) and our designed Density-K-Center algorithm, we explicitly identify representative structural anchors, establishing a more representative labeled domain. Second, to bridge the remaining distribution gap, we propose the Memory-guided Copy-Paste (MCP) module. Tailored for the inherent class imbalance in medical scans, MCP leverages a semantic memory mechanism to retrieve historically consistent priors for cross-domain alignment, encouraging semantic consistency. Coupled with an easy-to-hard progressive schedule, this framework effectively mitigates early-stage pseudo-label noise. Extensive experiments on six diverse 2D and 3D datasets demonstrate strong segmentation performance, particularly in extremely low-labeled scenarios (\eg, 1/16 ratio).

Problem

• semi-supervised medical image segmentation
• distribution bias
• low-data regime
• representative sampling
• medical data heterogeneity

Innovation

• Distribution-Aware Sampling
• Vision Foundation Models
• Memory-guided Copy-Paste
• Semi-Supervised Segmentation
• Class Imbalance