A CLUSTERING APPROACH TO CONSTRUCT MULTI-SCALE OVERCOMPLETE DICTIONARIES FOR ECG MODELING

The electrocardiogram (ECG) is the main biomedical signal used to diagnose and monitor cardiac pathologies. A typical ECG is composed of quasi-periodic activations (the QRS complexes, and the P and T waves) and periods of inactivity, plus noise and interferences. The sparse nature of the ECG has lead to the development of many compressed sensing (CS) and sparsity-aware ECG signal processing algorithms. In order to attain a good performance, these methods require appropriate dictionaries, and several online dictionary construction approaches have been devised. However, all of them require a substantial computational cost and the derived dictionaries are composed of atoms which may not be representative of real-world signals. In this work, we describe an efficient method for off-line construction of an overcomplete and multi-scale dictionary using a clustering-based approach. The resulting dictionary, whose atoms are the most representative waveforms from the training set, is then used to obtain a sparse representation of the ECG signal. Simulations on real-world records from Physionet's PTB database show the good performance of the proposed approach.