Automatic generation of realistic cardiopulmonary exercise test data with a conditional generative adversarial neural network

Abstract

This manuscript deals with the development of a generative model trained to create realistic cardiopulmonary test data. The model consists in a conditional generative adversarial neural network. The model can be used to generate an infinite number of fake-but-realistic cardiopulmonary tests. The gener-ated samples can be used in a variety of different applications, such as: 1) the creation of a shared dataset of well-defined tests that can be used to develop new methods for diagnostics and interpretation or to train exercise physiologists, and 2) to fill data gaps where artifacts are dominating the cardiopulmonary variables or when real data points are missing. The algorithm is deployed on a server and a web app can be used to challenge users to differentiate between a fake and a real test.