Privacy-Preserving Generative Deep Neural Networks Support Clinical Data Sharing

By Brett K. Beaulieu-Jones, Zhiwei Steven Wu, Christopher J. Williams, Ran Lee, Sanjeev P. Bhavnani, James Brian Byrd, Casey S. Greene. 2017

468 citations recorded by the source corpus.

Background Data sharing accelerates scientific progress but sharing individual level data while preserving patient privacy presents a barrier. Methods and Results Using pairs of deep neural networks, we generated simulated, synthetic “participants” that closely resemble participants of the SPRINT trial. We showed that such paired networks can be trained with differential privacy, a formal privacy framework that limits the likelihood that queries of the synthetic participants’ data could identify a real a participant in the trial. Machine-learning predictors built on the synthetic population generalize to the original dataset. This finding suggests that the synthetic data can be shared with others, enabling them to perform hypothesis-generating analyses as though they had the original trial data. Conclusions Deep neural networks that generate synthetic participants facilitate secondary analyses and reproducible investigation of clinical datasets by enhancing data sharing while preserving participant privacy.

Subfields: Artificial Intelligence, Health Informatics, Generative Models, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Health Informatics

In Generative Models

In Safety, Robustness & Interpretability

Citations in the corpus

Generative Models for Effective ML on Private, Decentralized Datasets (2019, background): “Prior work has looked at training GANs with DP in a centralized setting (Torkzadehmahani et al., 2019; Xie et al., 2018; Zhang et al., 2018; Frigerio et al., 2019; Beaulieu-Jones et al., 2018; Esteban et al., 2017).”
A survey of data fusion in smart city applications (2019, methodology): “In addition, we can leverage machine learning approaches such as [243], [244] to generate synthetic datasets with identical data characteristic for study purpose.”
Synthesizing Tabular Data using Generative Adversarial Networks (2018, background): “RGAN and RCGAN [13] can generate real-valued time-series data. medGAN [9], corrGAN [33] and several improved models [2, 6, 47, 3, 44] can generate discrete medical records but do not tackle the complexity in generating multimodal continuous variables. ehrGAN [8] generates augmented medical records…”
LOGAN: Membership Inference Attacks Against Generative Models (2017, methodology): “[6] apply the noisy gradient descent from [1] to train the discriminator of a Generative Adversarial Network under differential privacy.”
Real-valued (Medical) Time Series Generation with Recurrent Conditional GANs (2017, methodology): “Inspired by a recent preprint (Beaulieu-Jones et al., 2017), we apply the differential private stochastic gradient descent (DP-SGD) algorithm of (Abadi et al.”
Privacy-Preserving Distributed Deep Learning for Clinical Data (2018, background): “In particular, the addition of rich sources of patient data such as genomic measurements or retinal images make safe deidentiﬁcation extremely difﬁcult without signiﬁcantly compromising data utility [Beaulieu-Jones et al., 2017].”
Differentially Private Mixture of Generative Neural Networks (2017, methodology): “[5] apply the noisy gradient descent from [1] to train the discriminator of a Generative Adversarial Network [22] under differential privacy.”
Generative Adversarial Networks for Electronic Health Records: A Framework for Exploring and Evaluating Methods for Predicting Drug-Induced Laboratory Test Trajectories (2017, methodology): “They have been identiﬁed as a promising method to release de-identiﬁed biomedical data that can be shared while preserving privacy [1, 2] and can support model learning in supervised setups [4].”
Differentially Private Generative Adversarial Network (2018, background): “…privacy Zhao et al. [36] introduces energy-based GAN (EBGAN), which Berthelot et al. [4] also use an autoencoder as a discriminator et al. [3] used differentially private version of Auxiliary Classifier Choi et al. [8] proposed medGAN, which is Differential Privacy. include Dwork et al.…”
GAN-Leaks: A Taxonomy of Membership Inference Attacks against Generative Models (2019, background): “The generative model, on the other hand, has also gained increasing attention and has delivered appealing applications including photorealistic image synthesis [20, 44, 52, 70], text and sound generation [4, 49, 64, 65], sanitized dataset generation [2, 7, 35, 66, 73], etc.”