发表机构
Mohamed bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
PhaseAT提出一种相位感知对抗训练框架，通过在傅里叶域扰动相位并保持幅度不变，实现医学图像域泛化，在单源DG上提升超20%。
AI 中文摘要
深度医学图像模型的可靠临床部署受到跨扫描仪、站点和采集协议的分布偏移的阻碍。现有的域泛化（DG）方法通常侧重于风格或强度多样化，但它们仍可能使网络依赖于特定于域的纹理相关性。受傅里叶相位编码语义结构这一证据的启发，我们引入了PhaseAT，一种用于医学DG的相位感知对抗训练框架。PhaseAT通过在傅里叶域中迭代更新有界相位扰动同时保持幅度谱不变来形成相位扰动的训练视图，从而在匹配的外观统计下强调空间组织。扰动仅应用于YCbCr颜色空间中的亮度通道，以避免色度伪影。此外，一个简单的相位显著性掩码将更新集中在最有影响力的频率上。模型使用干净样本和相位扰动样本上的加权损失组合进行训练，支持单源和多源DG。我们在两个具有挑战性的医学数据集上验证了我们的方法，并证明PhaseAT在单源域泛化中实现了超过20%的提升，优于几种最先进的DG方法。代码实现可在以下网址获取：this https URL。
英文摘要
Reliable clinical deployment of deep medical image models is hindered by distribution shifts across scanners, sites, and acquisition protocols. Existing domain generalization (DG) methods often focus on style or intensity diversification, but they can still leave networks dependent on domain-specific texture correlations. Inspired by evidence that Fourier phase encodes semantic structure, we introduce PhaseAT, a phase-aware adversarial training framework for medical DG. PhaseAT forms phase-perturbed training views in the Fourier domain by iteratively updating a bounded phase perturbation while keeping the amplitude spectrum unchanged, thereby stressing spatial organization under matched appearance statistics. Perturbations are applied only to the luminance channel in YCbCr color space to avoid chromatic artifacts. Additionally, a simple phase-saliency mask concentrates updates on the most influential frequencies. The model is trained with a weighted combination of losses on clean and phase-perturbed samples, supporting both single-source and multi-source DG. We validate our method on two challenging medical datasets and demonstrate that PhaseAT achieves over 20% improvement in single-source domain generalization, outperforming several state-of-the-art DG methods. The code implementation is available at: https://github.com/ahmed-sharshar/PhaseAT.
CommentsThe paper is accepted in MICCAI 2026
Journal refMedical Image Computing and Computer Assisted Intervention - MICCAI 2026, Lecture Notes in Computer Science, vol. 16881, pp. 413-423, Springer, 2027
DOI：10.1007/978-3-032-38072-2_40
