Nonlinear dynamics analysis of electrocardiograms for detection of coronary artery disease

A computerized approach of nonlinear dynamics analysis of electrocardiogram (ECG) signals was applied for the detection of coronary artery disease (CAD). The proposed nonlinear dynamics descriptors were derived from 12-lead rest ECG data, and evaluated by originally developed computer software. Fluctuations of potentials of ECG leads that occur during the period of 20 ms with a magnitude of 5-20 mu V were significantly less beat-to-beat predictable in ischemic versus non-ischemic patients. The well-known nonlinear dynamics descriptors, recurrences percentage, mutual information, fractal dimension, and a new descriptor, next embedding dimension error, were good quantitative descriptors of fluctuations. They were significantly different (<p = 0.00001) in males with (108 patients) and without (54 patients) coronary artery lesions. The analysis of small fluctuations required a careful preprocessing technique based on knowledge of specifics of measurement errors and physiology of ECG signals. We considered finite differences of measured potentials with the time step of 20 ms as the initial source for nonlinear analysis. In nonlinear dynamics analysis, we also included such time moments that only belong to P- and T-waves or baseline drift with small positive slopes that allowed us to extract, under normal conditions, initial halves of P- and T-waves that displayed a better capacity to classify ischemic patients. (C) 2008 Elsevier Ireland Ltd. All rights reserved.