The Electrocardiogram (ECG) is a record of the electrical activity of the heart which serves as the first step in diagnosis of a cardiac abnormality. To detect and associate specific features linked to diseases is the thrust of research, aimed at the development of automated diagnostic tools. However, most of the existing studies use single-lead ECG data (from a single electrode), and focus only on diseases such as Arrhythmias and Chronic Heart Failure. The attempts to analyze full 12-lead ECG have been few, because of the short duration of data (few minutes) and lack of extensive training data that Machine Learning approaches require. Moreover, these approaches reveal little to no information of how a disease actually affects dynamics of the cardiac system. Based on Dynamical systems theory, studies on the nature and extent of chaos in cardiac systems already exist, and we use that insight to construct Multiplex Recurrence Networks (MRNs) from multi-lead ECG. These MRNs highlight spatio-temporal features of the cardiac dynamics which are quantified using layer similarity/dissimilarity measures in addition to the standard complex network measures. Based on the analysis of patient data from Bundle Branch Block, Dysrhythmia, Myocardial Infarction and Cardiomyopathy, we show that the cardiac dynamics manifests abnormalities in a multitude of ways that can be understood best with a set of measures that quantify different levels of structural complexity in the MRN. These results can lead to better classification and diagnosis algorithms that outperform existing ones. Since the framework developed can be used for any multivariate data, it may find applications outside of physiological data analysis.
