Refined composite multivariate multiscale entropy based on variance for analysis of resting-state magnetoencephalograms in Alzheimer's disease

Abstract

Alzheimer's disease (AD) is one of the fastest growing neurological diseases. Multiscale entropy with coarse-graining based on mean (MSEμ) has been widely used to characterize AD. Alternatively, multiscale entropy based on variance (MSEσ²) has been recently proposed to quantify the dynamics of volatility (variance) of univariate signals. Here, we extend the MSEσ² to multivariate signals to take into account both the time and spatial domains for discrimination of resting-state magnetoencephalogram (MEG) recordings of 36 AD patients from those of 26 normal controls. We also consider the usefulness of the refined composite mvMSEσ² (RCmvMSEσ²) to understand if the RCmvMSEσ² can better discriminate AD group from control subjects in comparison with mvMSEσ². The results show mvMSEσ² and RCmvMSEσ², unlike exiting multiscale-based methods, lead to significant differences between control and AD patients at all scale factors. The results obtained by the mvMSEσ² and RCmvMSEσ² are similar. Thus, refined composite technique might not enhance the detection of different pathological states, especially when signals are not too noisy and short. Finally, our findings show that the mvMSEσ² and RCmvMSEσ² can be useful tools for the analysis of real signals to characterize different kinds of dynamics.