Construction of wavelet dictionaries for ECG modelling

TL;DR

This paper details the development of various wavelet dictionaries to improve sparse ECG signal modeling, enhancing dimensionality reduction and signal representation efficiency.

Contribution

It introduces methods and algorithms for constructing multiple wavelet dictionary families, with software tools for easy extension and application to ECG sparse modeling.

Findings

Sparsity in ECG representation improves with the proposed wavelet dictionaries.

The approach outperforms standard wavelet bases in ECG sparse modeling.

Applicability demonstrated on multiple wavelet families.

Abstract

The purpose of sparse modelling of ECG signals is to represent an ECG record, given by sample points, as a linear combination of as few elementary components as possible. This can be achieved by creating a redundant set, called a dictionary, from where the elementary components are selected. The success in sparsely representing an ECG record depends on the nature of the dictionary being considered. In this paper we focus on the construction of different families of wavelet dictionaries, which are appropriate for the purpose of reducing dimensionality of ECG signals through sparse representation modelling. The suitability of wavelet dictionaries for ECG modelling, applying the Optimized Orthogonal Matching Pursuit approach for the selection process, was demonstrated in a previous work on the MIT-BIH Arrhythmia database consisting of 48 records each of which of 30 min length. This paper complements the previous one by presenting the technical details, methods, algorithms, and MATLAB software facilitating the construction of different families of wavelet dictionaries. The implementation allows for straightforward further extensions to include additional wavelet families. The sparsity in the representation of an ECG record significantly improves in relation to the sparsity produced by the corresponding wavelet basis. This result holds true for the 17 wavelet families considered here.