The feasibility of atrial and ventricular arrhythmias recognition using metrics of signal complexity for heartbeat intervals
It is well known that the cardiac system is controlled by the complex nonlinear self-regulation and the heartbeat variability (HRV) is an independent indicator of the autonomic regulation. With the assumption that intrinsic differences of atrial fibrillation (A) and ventricular ectopic arrhythmias (V) can be unveiled by a proper approach based on of signal complexity, we examine the feasibility of detecting these arrhythmias of different pathological origins using metrics of complexity for heartbeat intervals (HRI). Specifically, the normal sinus rhythm (N), the A type and the V type are used as the targeted types of heartbeat. By extracting the entropy-based features from HRI of different lengths, i.e., from 300 heartbeats to 1000 heartbeats, we examined the distinguishability of these 3 types of heartbeat. By applying the features to the random forest model, the HRI signal of 600-heartbeat-length can be used to detect the A and V completely, i.e., with 100% of type-wise recall and precision. What is more, this approach is sensitive to the existence of the corresponding arrhythmias. The results substantiate our assumption about the intrinsic difference of the A and V type. A further investigation applying this approach to a wider spectrum and a finer stratification of arrhythmias/ cardiac diseases and may lead to the systematic understanding in the context of complexity and better insight for its practical use for wearable/unconstrained monitors.