发表机构
Northwestern University; University of Wisconsin-Madison; Istanbul University; Namik Kemal University; Istanbul Florence Nightingale Hospital; Acibadem Mehmet Ali Aydinlar University; Qatar University; Boston University; Mohamed bin Zayed University of Artificial Intelligence; NVIDIA（西北大学; 威斯康星大学麦迪逊分校; 伊斯坦布尔大学; 纳米克·凯末尔大学; 伊斯坦布尔佛罗伦萨南丁格尔医院; 阿西巴德姆·穆罕默德·阿里·艾丁拉尔大学; 卡塔尔大学; 波士顿大学; 穆罕默德·本·扎耶德人工智能大学; 英伟达公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究构建BreastMammo和DenseMammo乳腺钼靶数据集，提出含仅前景直方图匹配协议的领域泛化框架，在密度分类任务中实现高AUC，且跨数据集泛化性能优于主流范式。
AI 中文摘要
乳腺密度分类是乳腺癌风险评估的关键组成部分，但AI模型常因不同医疗机构的设备采集风格差异而难以跨站点泛化。本研究引入两个新数据集BreastMammo和DenseMammo，以推动稳健的多视角乳腺钼靶研究。我们提出一种领域泛化框架，采用仅前景的直方图匹配协议解决不同临床来源产生的领域偏移问题。使用5折交叉验证协议进行内部评估，结果显示Swin Transformer主干网络在密度分类任务中达到峰值AUC 98.32%；在TNMammo和LUMINA数据集上的外部评估表明，该方法能持续减少领域偏移，显著优于MixStyle及基于离散傅里叶变换的框架等主流领域泛化范式。
英文摘要
Breast density classification is a critical component of breast cancer risk assessment, yet AI models often struggle to generalize across clinical sites due to vendor-specific acquisition styles. In this work, we introduce two new datasets, BreastMammo and DenseMammo, to facilitate robust multi-view mammography research. We propose a domain generalization framework that utilizes a foreground-only histogram matching protocol to resolve the domain shift issue arising from disparate clinical sources. Internal evaluation using a 5-fold cross-validation protocol demonstrates the efficacy of our approach, with the Swin Transformer backbone achieving a peak AUC of 98.32% for density classification. External evaluation on the TNMammo and LUMINA datasets demonstrates that the proposed approach consistently reduces domain shift, significantly outperforming prominent domain generalization paradigms, including MixStyle and Discrete-Fourier-Transform-based frameworks.
CommentsThis paper was accepted to the MICCAI 2026 workshop Deep-Brea3th
