arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.03264cs.CVcs.LGeess.IV
收藏

HypCBC: 领域不变的双分支一致性用于通用化医学图像分析

HypCBC: Domain-Invariant Hyperbolic Cross-Branch Consistency for Generalizable Medical Image Analysis

Bamberg University of Bamberg（巴马克伯格大学）；xAILab（xAIL实验室）

Francesco Di Salvo, Sebastian Doerrich, Jonas Alle, Christian Ledig

阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

HypCBC通过双曲流形建模提升医学图像分析的领域不变性，实现跨不同数据集和模态的泛化性能提升。

摘要：

深度神经网络在训练分布之外的鲁棒泛化仍是一个关键挑战。这在医学图像分析中尤为突出，因为数据通常稀缺，且由于不同的硬件设备、成像协议和异质患者群体，存在协变量偏移。这些因素共同阻碍了可靠的性能并减缓了临床应用。尽管最近有进展，现有的学习范式主要依赖于欧几里得流形，其平坦的几何结构无法捕捉临床数据中存在的复杂、分层结构。在本工作中，我们利用双曲流形的优势来建模复杂的数据特征。我们首次全面验证了双曲表示学习在医学图像分析中的应用，并在11个分布内数据集和3个ViT模型上展示了统计显著的提升。我们进一步提出了一个无监督、领域不变的双曲交叉分支一致性约束。大量实验证实，我们提出的方法促进了领域不变的特征，并在三个领域泛化基准上比最先进的欧几里得方法平均提高了+2.1%的AUC：Fitzpatrick17k、Camelyon17-WILDS和视网膜成像的跨数据集设置。这些数据集涵盖不同的成像模态、数据大小和标签粒度，证实了在显著不同的条件下泛化能力。代码可在https://github.com/francescodisalvo05/hyperbolic-cross-branch-consistency获取。

英文摘要：

Robust generalization beyond training distributions remains a critical challenge for deep neural networks. This is especially pronounced in medical image analysis, where data is often scarce and covariate shifts arise from different hardware devices, imaging protocols, and heterogeneous patient populations. These factors collectively hinder reliable performance and slow down clinical adoption. Despite recent progress, existing learning paradigms primarily rely on the Euclidean manifold, whose flat geometry fails to capture the complex, hierarchical structures present in clinical data. In this work, we exploit the advantages of hyperbolic manifolds to model complex data characteristics. We present the first comprehensive validation of hyperbolic representation learning for medical image analysis and demonstrate statistically significant gains across eleven in-distribution datasets and three ViT models. We further propose an unsupervised, domain-invariant hyperbolic cross-branch consistency constraint. Extensive experiments confirm that our proposed method promotes domain-invariant features and outperforms state-of-the-art Euclidean methods by an average of $+2.1\%$ AUC on three domain generalization benchmarks: Fitzpatrick17k, Camelyon17-WILDS, and a cross-dataset setup for retinal imaging. These datasets span different imaging modalities, data sizes, and label granularities, confirming generalization capabilities across substantially different conditions. The code is available at https://github.com/francescodisalvo05/hyperbolic-cross-branch-consistency .

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。

↑