Coarse-Graining Approaches in Univariate Multiscale Sample and Dispersion Entropy

Area of Science:

• Signal processing
• Complexity science
• Biomedical engineering

Background:

• Multiscale Entropy (ME) is a key method for evaluating signal complexity across different time scales.
• Classical ME involves coarse-graining and entropy estimation, with recent focus on short time series analysis.
• The impact of downsampling within coarse-graining and its relation to filtering methods remains underexplored.

Purpose of the Study:

• To systematically assess the influence of coarse-graining downsampling on Multiscale Entropy (ME) estimations.
• To compare classical moving average downsampling with low-pass Butterworth filtering and empirical mode decomposition.
• To evaluate these methods using synthetic and real physiological data.

Main Methods:

• Comparison of Sample Entropy and Dispersion Entropy with different coarse-graining strategies.
• Implementation of moving average and Butterworth low-pass filtering, with and without downsampling.
• Analysis using Intrinsic Multiscale Entropy with empirical mode decomposition on synthetic and physiological datasets.

Main Results:

• Downsampling significantly impacts entropy values, decreasing them at low sampling frequencies and increasing them at high frequencies.
• The refine composite method offers minimal improvement for long signals with low noise at a high computational cost.
• Downsampling is often not required for accurate signal complexity quantification, particularly for shorter time series.

Conclusions:

• Downsampling's role in coarse-graining for ME requires careful consideration and may not always be necessary.
• Findings support the development of more robust, efficient, and noise-resistant ME techniques for diverse recording lengths.
• This research contributes to optimizing ME analysis for both short and long signal recordings.