Table 2.
Summary of approximate entropy, sample entropy, wavelet entropy and multiscale entropy.
| Entropy | Interpretation | Advantages | Limitations |
|---|---|---|---|
| Approximate Entropy (ApEn) | The larger the ApEn, the less the predictability or the higher the randomness | 1. ApEn correlates with hidden and subclinical changes often undetected by other classical time series analysis (moment statistics, spectral analysis, and correlation analysis) 2. ApEn can assess subtle disruption, typically preceding change in signal mean and standard deviation |
1. The higher entropy value only indicates an increase in the degree of randomness rather than complexity 2. The calculation usually require very long data sets and a bias may exist leading to overestimation of the time series regularity |
| Sample Entropy (SampEn) | The larger the SampEn, the less the predictability or the higher the randomness The larger value of SamEn, the less self-similarity | 1. Simpler than ApEn 2. Largely independent of record length and thus consistency 3. Less biased than ApEn since it eliminates self-matches |
1. The estimation of SampEn critically depends on the selection of the parameters' sequence length 2. The stationarity assumption is invalid for prolonged time periods |
| 1. Wavelet Entropy (WEn) 2. Relative Wavelet Entropy (rWEn) | The larger the wavelet entropy, the less the predictability or the higher the randomness | 1. Wavelet entropy has similar performance with ApEn 2. Inherits the high computational efficiency of wavelet decomposition 3. rWEn could be further used to measure dissimilarity between two time series signals |
1. The higher entropy value only indicates an increase in the degree of randomness rather than complexity 2. Parameter selection in wavelet decomposition could cause bias in clinical practice |
| Multiscale Entropies (MSE) | The larger the multiscale entropy, the increase degree of complexity | 1. Characterize complexity in signal better than other entropies 2. Multiscale entropy can attenuate the effect of the stationarity assumption in the underlying distribution of signals |
Multiscale entropy requires substantially more samples than single scale sample entropy |