[Automatic Classification of Epileptic Electroencephalogram Signal Based on Improved Multivariate Multiscale Entropy]

Area of Science:

• Nonlinear dynamics
• Biomedical signal processing
• Complexity science

Background:

• Traditional sample entropy struggles with long-range dependencies in real-world data.
• Multiscale sample entropy (MSE) is effective for univariate data but limited for multichannel systems.
• Existing multivariate multiscale entropy methods are computationally intensive and time-consuming.

Purpose of the Study:

• To develop a more efficient and accurate multivariate multiscale entropy method for multichannel signal analysis.
• To address the computational and memory limitations of traditional multivariate methods.
• To improve the timely and accurate reflection of nonlinear dynamic correlations in multivariate signals.

Main Methods:

• An improved multivariate multiscale entropy method was developed, embedding on all variables simultaneously.
• This novel embedding strategy overcomes memory overflow issues with increasing channel numbers.
• The method was validated using simulation data and the Bonn epilepsy dataset.

Main Results:

• Simulation results demonstrated the method's capability in distinguishing correlated data.
• The improved method achieved superior classification accuracy on the Bonn epilepsy dataset.
• Specifically, 100% accuracy was obtained for the Z and S data collections.

Conclusions:

• The proposed improved multivariate multiscale entropy offers a computationally efficient and accurate solution for analyzing complex multichannel signals.
• This method is well-suited for real-world multivariate signal analysis, including applications in epilepsy detection.
• The enhanced approach provides a valuable tool for understanding nonlinear dynamic correlations in complex systems.