arXiv：cs.AI· Soroosh Tayebi Arasteh, Mina Farajiamiri, Mahshad Lotfinia, Behrus Hinrichs-Puladi, Jonas Bienzeisler, Mohamed Alhaskir, Mirabela Rusu, Christiane Kuhl, Sven Nebelung, Daniel Truhn·· 18 小时前
预训练领域比训练目标更能决定差分隐私医学影像分析的隐私-效用权衡
The pretraining domain outweighs the training objective in setting the privacy-utility trade-off of differentially private medical image analysis
AI 导读
在591,884张胸片、5个数据集上从5种起点训练差分隐私胸片分类器，有监督胸片预训练在25种数据集与隐私设置组合中24种排名第一，其AUC在无隐私时领先自然图像有监督预训练2.6%，在最严格预算下领先15.3%。改变预训练领域对该AUC的提升是改变目标的2.2至3.7倍；隐私下的胸片预训练在最严格预算下AUC约降5%，但仍优于三种无患者数据的起点。
正文
Abstract:Differential privacy limits how much each training image can influence a medical imaging model but lowers its diagnostic accuracy. Pretrained weights reduce this loss. It is unknown whether the pretraining objective or domain improves accuracy more. Here we train chest radiograph classifiers under differential privacy from five starting points on 591,884 radiographs from five datasets. We show that supervised pretraining on chest radiographs ranks first in 24 of 25 combinations of dataset and privacy setting. Measured by the area under the receiver operating characteristic curve, its lead over supervised pretraining on natural images is 2.6% without privacy and 15.3% at the strictest budget. From that starting point, changing the domain improves this area 2.2 to 3.7 times more than changing the objective. Private pretraining on chest radiographs lowers this area by about 5% under privacy but outperforms the three starting points without patient data at the strictest budget. Under privacy, supervised chest radiograph pretraining leads in the worst-performing demographic subgroup.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2601.19618 [cs.CV] |
| (or arXiv:2601.19618v3 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2601.19618 arXiv-issued DOI via DataCite |
Submission history
From: Soroosh Tayebi Arasteh [view email]
[v1]
Tue, 27 Jan 2026 13:50:43 UTC (1,781 KB)
[v2]
Wed, 22 Jul 2026 19:51:14 UTC (430 KB)
[v3]
Sat, 10 Oct 2026 13:15:01 UTC (474 KB)
来源：arXiv：cs.AI · arxiv.org