The automatic and rapid analysis of long-term electrocardiogram (ECG) records still remains a challenging task. Most of the existing algorithms are time consuming and require a training step. In this paper, we present a training free two-level hierarchical model based on ordinal patterns for classifying ECG beats into three types. The classification rules include morphological and temporal properties of the ECG signal that are compared to RR and QRS dependent thresholds derived from the beat CEOP or PE series. The experimental classification rates obtained from the MIT-BIH Arrhythmia database (93.66%) and the St. Petersburg Institute of Cardiological Technics (INCART) database (95.43%), considering the Advancement of Medical Instrumentation (AAMI) recommendations, confirm the ability of the proposed approach for a multi-class classification.

In this paper, we investigate the applicability of the permutation entropy (PE) and the conditional entropy of ordinal patterns (CEOP) to Electrocardiogram (ECG) data analysis. We define a signal dependent threshold based on the PE and the CEOP for the detection of abnormal ECG beats. Parameters of the proposed threshold formula are calibrated using the MIT-BIH Arrhythmia and the European Society of Cardiology ST-T (ESC) databases. The experimental results show that the difference between CEOP and PE is marginal, and that the algorithm is less sensitive to the parameter setting. We achieved a classification rate of 93.62% in the case of the MIT-BIH database, and 99.57% in the case of the ESC database. Although these algorithms still need to be improved, the above results for the ESC database confirm that ordinal pattern based entropies are promising for ECG beat classification. (C) 2019 Elsevier B.V. All rights reserved.