自监督预训练在差分隐私医疗图像分析中的作用
The pretraining domain outweighs the training objective in setting the privacy-utility trade-off of differentially private medical image analysis
AI总结：
本文研究了自监督预训练在差分隐私医疗图像分析中的作用，发现DINOv3初始化在DP下优于ImageNet初始化，但不如领域特定监督预训练。
摘要：
差分隐私（DP）为敏感数据提供了正式的保护，但通常会带来诊断性能显著的损失。模型初始化已成为缓解这种退化的关键因素，然而现代自监督学习在全模型DP中的作用仍不明确。本文通过大规模评估初始化策略，针对差分隐私医疗图像分析进行研究，以胸片分类为代表性基准，使用超过800,000张图像。使用最先进的ConvNeXt模型在现实隐私范围内训练DP-SGD，我们比较了非领域特定的监督ImageNet初始化、非领域特定的自监督DINOv3初始化以及领域特定的监督预训练（在MIMIC-CXR数据集上）。评估在五个外部数据集上进行，涵盖不同机构和采集设置。我们证明，在DP下，DINOv3初始化相比ImageNet初始化在诊断效用上持续提升，但仍然不如领域特定的监督预训练，后者在性能上最接近非隐私基线。我们进一步表明，初始化选择强烈影响人口公平性、跨数据集泛化能力以及在隐私约束下的数据规模和模型容量的鲁棒性。结果确立了初始化策略作为差分隐私医学成像中效用、公平性和泛化能力的关键决定因素。
英文摘要：
Differential privacy protects the patients whose images train medical imaging models, but it lowers diagnostic accuracy, and the initialization is the strongest known remedy. Practice increasingly favors large generic self-supervised encoders. Yet the pretraining objective and the pretraining domain are confounded in existing comparisons, so which one preserves utility under privacy is unknown, and the pretraining corpus is treated as public even when it holds patient images. We trained ConvNeXt classifiers with differentially private stochastic gradient descent from five initializations that vary the objective and the domain independently, at four privacy budgets and without privacy, and evaluated them locally on more than 590,000 chest radiographs from five external datasets in four countries. Supervised pretraining on chest radiographs ranked first in 24 of 25 dataset and budget combinations. Its lead over ImageNet grew from 2.5 to 14.6 points of macro-averaged area under the receiver operating characteristic curve as the budget tightened, and the domain effect exceeded the objective effect by a factor of 2.2 to 3.4. Pretraining that corpus privately cost about 5 points and, under privacy, still beat every public initialization. Low-rank adaptation removed about half the residual gap, and in-domain pretraining raised the worst-performing demographic subgroup. Under privacy, what a model was pretrained on outweighs how it was pretrained.
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。
