arXiv 2605.135442026-07-03cs.CV

CA-GCL：跨解剖全局-局部对比学习用于稳健的3D医学图像理解

CA-GCL: Cross-Anatomy Global-Local Contrastive Learning for Robust 3D Medical Image Understanding

University of Electronic Science and Technology of China（电子科技大学）；Mohamed bin Zayed University of Artificial Intelligence（莫扎德人工智能大学）

Hanwen Zhang, Yao Liu, Die Dai, Jiaye Yang, Qiao Liu, Yutong Xie, Peng Wang

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出CA-GCL框架，通过全局对比学习和临床感知文本增强，解决3D医学图像理解中文本嵌入空间退化问题，提升零样本异常检测性能和跨数据集泛化能力。

摘要：

细粒度视觉-语言预训练（FVLP）通过对齐解剖级视觉表示与相应文本描述，在3D医学图像理解中展现出巨大潜力。然而，现有FVLP范式常面临文本嵌入空间严重退化问题，不同解剖结构的文本嵌入高度聚类且难以区分。为此，本文提出新的跨解剖全局-局部对比学习框架（CA-GCL）。CA-GCL引入全局对比目标，强制潜在空间中解剖类别分离，有效对抗局部对齐诱导的聚合倾向。此外，我们基于置换不变性和部分完整性引入临床感知文本增强策略，以增强对描述不完整性的鲁棒性。在CT-RATE和Rad-ChestCT数据集上的广泛评估表明，CA-GCL在零样本异常检测中持续优于现有VLP范式，实现更优性能的同时表现出强跨数据集泛化能力。关键地，CA-GCL减少了不同提示模板间的性能方差，将退化的文本相似度分布转化为钟形分布。这些结果验证了CA-GCL作为稳健3D医学图像理解有效框架的可行性。

英文摘要：

Fine-grained Vision-Language Pre-training (FVLP) demonstrates significant potential in 3D medical image understanding by aligning anatomy-level visual representations with corresponding textual descriptions. However, existing FVLP paradigms often suffer from severe representation collapse in the textual embedding space, where text embeddings of distinct anatomical structures become highly clustered and indistinguishable. This distributional degeneracy renders the model hypersensitive to prompt variations, hindering reliable clinical deployment. To address these challenges, we propose a novel Cross-Anatomy Global-Local Contrastive Learning framework (CA-GCL). CA-GCL introduces a global contrastive objective that enforces separation between anatomical categories in the latent space, effectively counteracting the aggregation tendency induced by local alignment. Furthermore, we incorporate a clinical-aware text augmentation strategy based on permutation invariance and partial completeness to enhance robustness against descriptive incompleteness. Extensive evaluations on the CT-RATE and Rad-ChestCT datasets show that CA-GCL achieves comparable zero-shot abnormality detection performance to existing VLP paradigms, while demonstrating substantially better robustness to prompt variations: on canonical templates it obtains higher mean AUC with lower variance, and on non-canonical templates it remains stable whereas baselines degrade markedly. These results validate CA-GCL as an effective framework for robust 3D medical image understanding.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。