Synthesis of Dependent Multichannel ECG using Generative Adversarial Networks

Access to medical data is highly regulated due to its sensitive nature, which can constrain communities' ability to utilise these data for research or clinical purposes. Common de-identification techniques to enable the sharing of data may not provide adequate protection for an individual's personal data in every circumstance. We investigate the ability of Generative Adversarial Networks (GANs) to generate realistic medical time series data to address these privacy and identification concerns. We generate synthetic, and more significantly, multichannel electrocardiogram (ECG) signals that are representative of waveforms observed in patients. Successful generation of high-quality synthetic time series data has the potential to act as an effective substitute for actual patient data. For the first time, we demonstrate a multivariate GAN architecture that can successfully generate dependent multichannel time series signals. We present the first application of multivariate dynamic time warping as a means of evaluating generated GAN samples. Quantitative evidence demonstrates our GAN can generate data that is structurally similar to the training set and diverse across generated samples, all whilst ensuring sufficient privacy guarantees for the underlying training data.