Impact of compressed sensing on clinically relevant metrics for ambulatory ECG monitoring

Abstract

Recent research has examined the combination of compressed sensing with over-complete dictionaries for the lossy compression of electrocardiogram (ECG) signals. The application of dictionary learning to automatically create the dictionary is described. A novel analysis of the reconstructed signals using a range of clinical metrics based around QRS feature extraction and heart rate variability is employed. Two methods for dictionary creation are proposed: Patient specific and patient agnostic. A detailed comparison of each approach is described. Considering ambulatory ECG monitoring as an application, each methodology is analysed for a wide range of compression ratios.