Overcomplete Multi-scale Dictionaries for Efficient Representation of ECG Signals

Abstract

The electrocardiogram (ECG) was the first biomedical signal subject of extensive digital signal processing techniques. Essentially, the ECG consists of a cyclic sequence of relevant activations embedded into inactivity time sequences combined with interferences and noise. By its nature, it can be subject of representation as a sparse signal. This work describes an efficient method to create overcomplete multi-scale dictionaries that can be used for sparse ECG representation. Whereas most of the proposed methods to date use fixed waveforms that somehow resemble actual ECG shapes, the main innovation in our approach is selecting ECG waveforms recorded from actual patients. A relevant result of our method is the ability to process long lasting recordings from multiple patients. Simulations on patient actual records from Physionet's PTB Diagnostic ECG Database confirm the good performance of the proposed approach.