发表机构
College of Engineering and Computer Science, VinUniversity; Imperial College London; Radiology Department, Vietnam National Cancer Hospital; VinUni-Illinois Smart Health Center, VinUniversity; The Computer Vision and Medical AI Lab, VinUniversity（工程与计算机科学学院，文大大学; 伦敦帝国理工学院; 越南国家癌症医院放射科; 文大大学 - 伊利诺伊智能健康中心，文大大学; 计算机视觉与医学人工智能实验室，文大大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究在域转移下乳腺钼靶成像基础模型的稳健性，用统一协议在多数据集上训练评估15种模型主干，发现特定视觉语言模型性能强，DINOv3是有竞争力基线，适应预训练未持续提升泛化，强调数据集级OOD评估是核心标准。
AI 中文摘要
基础模型越来越多地被用作乳腺钼靶成像的图像特征提取器，但其在外部域转移下的稳健性仍不明确。我们使用统一的冻结主干线性探针协议，在3个源数据集上训练，并在标签协调后在12个任务兼容的分布外（OOD）数据集上评估，对15种基础模型主干在乳腺密度、BI-RADS严重程度和癌症状态方面进行基准测试。乳腺钼靶特定的视觉语言模型（Mammo-FM和MaMA）提供了最强的平均OOD性能，但稳健性不能仅由乳腺钼靶暴露来解释。DINOv3仍然是一个有竞争力的仅视觉基线，并且适应乳腺钼靶的预训练并不能持续提高泛化能力。数据集级分析进一步表明，即使是领先模型在不同数据集上的性能也存在异质性。特征空间检查表明，有用的表示可以在保留数据集和采集结构的同时保留临床信号。这些发现突出了数据集级OOD评估作为评估乳腺钼靶表示的核心标准。我们的代码可公开获取：此https URL。
英文摘要
Foundation models are increasingly used as image feature extractors for mammography, but their robustness under external domain shift remains unclear. We benchmark 15 foundation-model backbones across breast density, BI-RADS severity, and cancer status using a unified frozen-backbone linear-probe protocol, training on 3 source datasets and evaluating on 12 task-compatible out-of-distribution (OOD) datasets after label harmonization. Mammography-specific vision-language models (Mammo-FM and MaMA) provide the strongest mean OOD performance, but robustness is not explained by mammography exposure alone. DINOv3 remains a competitive vision-only baseline, and mammography-adapted pretraining does not consistently improve generalization. Dataset-level analysis further shows that even leading models show heterogeneous performance across datasets. Feature-space inspection reveals that useful representations can preserve clinical signal while retaining dataset and acquisition structure. These findings highlight dataset-level OOD evaluation as a central criterion for assessing mammography representations. Our code is publicly available: https://github.com/biomedia-mira/mammo-ood.
CommentsAccepted at Deep-Brea3th 2026 workshop in conjunction with MICCAI 2026
