Accurate ECG monitoring by Gaussian feature streaming

Abstract

Wearable cardiac monitors can usefully contribute to early detection of potential cardiovascular pathologies. However ECG trace data streaming over wireless links creates some significant challenges, due to the amount of data to be transmitted. We employ a signal analysis approach based on a Gaussian dictionary to model ECG traces in a compressed way. The algorithm operates on fixed-length segments, and achieves effective compression for wireless data transmission, associating just 6 bytes to each Gaussian feature. At the same time it enables accurate reconstruction of ECG traces from the reduced data set. We tested our method on a set of 46 ECG recordings taken from the Physionet MIT-BIH Arrhythmia Database, obtaining 90% data compression rates, while percent relative deviation of reconstructed traces is always below 5%.