Non-linear modeling analysis of the high-resolution ECG for estimating abnormal intra-QRS potentials

Presents a non-linear modeling analysis of the high-resolution electrocardiogram (HRECG) with the purpose of measuring abnormal intra-QRS potentials (AIQP). A non-linear autoregressive with an exogenous input (NARX) model structure, parametrized by a multilayer feedforward neural network, was used for estimating the smooth normal part of the QRS waveform. Each individual-lead HRECG is presented unfiltered to be mathematically modeled The modeling procedure was applied to 73 non-event subjects and 59 patients with ventricular tachycardia (VT) and high probability of intra-QRS signals after myocardial infarction. The technique is capable of separating relatively predictable (normal) and unpredictable (abnormal) components of the QRS of each individual HRECG lead. Mean AIQP values are significantly greater in VT group in all three leads: p<0.01 for leads X and Z; p<0.05 for lead Y. The non-linear modeling technique can improve the characterization of abnormal signals within the QRS complex for detecting patients with arrhythmic events.