arXiv 2604.002632026-03-31eess.IVcs.CV

多机构级特征泄露减少用于跨医院胸部X光转移的自监督学习

Feature-level Site Leakage Reduction for Cross-Hospital Chest X-ray Transfer via Self-Supervised Learning

University of Mentouri Brothers Constantine 1（君士坦丁兄弟门图里大学）；LRIA Laboratory, University of Science and Technology Houari Boumediene (USTHB)（LRIA实验室，胡阿里·布迈丁科技大学）；LSIACIO Laboratory, University of Mentouri Brothers Constantine 1（LSIACIO实验室，君士坦丁兄弟门图里大学）；University of Paris-VIII-Vincennes-Saint-Denis（巴黎第八大学）

Ayoub Louaye Bouaziz, Lokmane Chebouba

备注：: Accepted at The 7th International Conference on Computing Systems and Applications [Algiers,2026]

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文研究了如何直接测量多机构泄露，并探讨其对跨医院转移方法的影响。通过多机构自监督学习和特征级对抗混淆，改进了胸部X光转移的性能，发现多机构SSL提升AUC，而对抗混淆降低泄露但不显著提升性能。

摘要：

跨医院胸部X光模型的失败常归因于域移位，但大多数工作假设不变性而未测量。本文研究如何直接测量机构泄露并探讨其对转移方法的影响。我们研究多机构自监督学习（SSL）和特征级对抗混淆用于跨医院转移。我们先在NIH和CheXpert上预训练ResNet-18，不使用病理标签。然后冻结编码器，仅在NIH上训练线性肺炎分类器，评估转移到RSNA的效果。我们使用后验线性探针量化机构泄露，该探针通过冻结骨干特征f和投影特征z预测获取机构。在3个随机种子下，多机构SSL将RSNA AUC从0.6736±0.0148（ImageNet初始化）提升至0.7804±0.0197。在f上添加对抗混淆降低测量泄露但不可靠提升AUC并增加方差。在f上，机构探针准确率从0.9890±0.0021（SSL-only）降至0.8504±0.0051（CanonicalF），偶然率为0.50。在z上，探针准确率从0.8912±0.0092降至0.7810±0.0250。这些结果表明测量泄露改变了转移方法的解释：多机构SSL驱动转移，而对抗混淆暴露了不变性假设的局限性。

英文摘要：

Cross-hospital failure in chest X-ray models is often attributed to domain shift, yet most work assumes invariance without measuring it. This paper studies how to measure site leakage directly and how that measurement changes conclusions about transfer methods. We study multi-site self-supervised learning (SSL) and feature-level adversarial site confusion for cross-hospital transfer. We pretrain a ResNet-18 on NIH and CheXpert without pathology labels. We then freeze the encoder and train a linear pneumonia classifier on NIH only, evaluating transfer to RSNA. We quantify site leakage using a post hoc linear probe that predicts acquisition site from frozen backbone features $f$ and projection features $z$. Across 3 random seeds, multi-site SSL improves RSNA AUC from 0.6736 $\pm$ 0.0148 (ImageNet initialization) to 0.7804 $\pm$ 0.0197. Adding adversarial site confusion on $f$ reduces measured leakage but does not reliably improve AUC and increases variance. On $f$, site probe accuracy drops from 0.9890 $\pm$ 0.0021 (SSL-only) to 0.8504 $\pm$ 0.0051 (CanonicalF), where chance is 0.50. On $z$, probe accuracy drops from 0.8912 $\pm$ 0.0092 to 0.7810 $\pm$ 0.0250. These results show that measuring leakage changes how transfer methods should be interpreted: multi-site SSL drives transfer, while adversarial confusion exposes the limits of invariance assumptions.

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。