As we move from manned missions to the International Space Station (ISS) to more distant interplanetary travel to the moon, Mars, and beyond, humans are exposed for longer periods of time to radiation and microgravity. Under these conditions, astronauts undergo numerous physiological changes which are not yet fully understood. In particular, the heart experiences atrophy combined with increased risks of cardiovascular anomalies such as atrial fibrillation.
To understand, monitor, and potentially diagnose heart conditions while astronauts are in space, space agencies are utilizing wearable devices to collect continuous biosignals of their astronauts. Data from this device enables the creation of a real-time cardiac monitoring and diagnostic tool for astronauts. However, there is a lack of astronaut health data which is essential for training any effective, autonomous system capable of detecting and diagnosing medical conditions in future missions.
Machine learning, specifically generative models, enable us to address this lack of data - by synthesizing symptomatic data that resembles the output of a wearable device. We use a deep learning approach, translating the symptomatic morphology and domain specific characteristics of an ECG signal into style and content, akin to concepts expressed in the domain of computer vision. The ECG Generator of Representative Encoding of Style and Symptoms (EGRESS) is a model with the potential to generate millions of ECG symptomatic signals, an important step in working to build an astronaut health monitoring system for space agencies worldwide. The work presented here is a first step towards generating various classes of ECG data for the purpose of training future health monitoring systems in space and advancing research in the wider biosignal domain.