Suitability of multiscale entropy for complexity quantification of cardiac rhythms in chronic pathological conditions: a similarity patterns based investigation

In this paper, based upon the appearance of patterns derived from a time series, we have investigated the suitability of multiscale entropy (MSE) technique for complexity quantification of cardiac rhythms in chronic pathological conditions. MSE analysis was developed to quantify the complexity of a wide variety of biomedical signals. Here, sample entropy (SampEn) technique was evaluated across multiple spatio-temporal scales. In SampEn, to find the appearance of repetitive patterns in multi-dimensional phase space, the threshold value ‘s’ is pre-fixed as 0.2. However, the cardiac rhythms of some pathologies are characterized with considerable erratic beat-to-beat fluctuations, and hence, in accordance with that, the patterns concealed in the pathologic cardiac rhythms spread across a wider region of multidimensional phase space. But, fixed threshold value ‘s’ assigns a fewer similarity pattern inside a circle of fixed dimensions, and hence, the higher entropy rate is associated with the chronic pathologic cardiac rhythms when compared to healthy cardiac rhythms. This flaw of SampEn is present in MSE, which leads to the wrong estimation of complexity associated with a time series. The outcome of this issue is clearly visible at low time scales, where period-to-period fluctuations in chronic pathologic cardiac rhythms and in randomized time series are significantly increased. In this present study, MSE analysis was performed over synthetic simulated database comprising of (white noise) WN and (power noise) PN signals. Further, MSE analysis was performed on the RR-interval series collected from (normal sinus rhythm) NSR group, and patients affected by (Atrial Fibrillation) AF. A fixed number of data samples ‘M’ of 10,000 were considered for each type of time series. Here, it is being observed that at some time scales, MSE assigns higher entropy to the WN and AF group, rather than PN and NSR group respectively, which is a wrong estimation of complexity. However, both the groups are discriminated efficiently by this algorithm. Further, it is concluded that MSE measure both the entropy and short term variations associated with a time series, but unable to investigate the real complexity (meaning full structural organization) present in a signal.