CorGAN-Correlation Capturing Convolutional Neural Networks for Generating Synthetic Healthcare Records

• Amirsina Torfi  |  Edward A. Fox

• Virginia Polytechnic Institute and State University

In healthcare domain, researchers face a lot of privacy challenges when it comes to create a deep learning model using electronic health record (EHR) data. This drags attention towards generating realistic synthetic data with privacy ensured. To generate synthetic healthcare records an efficient architecture is proposed using Convolutional GANs and Convolutional Auto-encoders which is named as "CorGAN". It is shown that the Convolutional Neural Networks are more effective as compared to Multilayer Perceptrons to capture inter-correlation between features. It is demonstrated that CorGAN can generate realistic synthetic data that performs similar to real data on classification task. The datasets used for demonstration are MIMIC-III dataset and UCI Epileptic Seizure Recognition dataset. The effectiveness of proposed model is shown using Stacked Deep Boltzmann Machines (DBMs), Variational Autoencoders and medGan. A privacy assignment is also given and it is observed that the model provides an acceptable level of privacy, by varying the amount of synthetically generated data and amount of data known to an adversary.

Input variables : Source EHR, EEG
Output Variables : Synthetic EHR, Status of Epileptic Seizure

Visit Model : github.com

Additional links : arxiv.org