Performance analysis of four nonlinearity analysis methods using a model with variable complexity and application to...

Area of Science:

• Time Series Analysis
• Nonlinear Dynamics
• Biomedical Signal Processing

Background:

• Detecting nonlinear characteristics in time series is crucial for understanding complex systems.
• Existing nonlinear methods and surrogate data analysis (z-score) have limitations in sensitivity and robustness evaluation.
• The comparative performance of different nonlinear methods, especially against linear techniques in real-world applications, requires further investigation.

Purpose of the Study:

• To investigate and quantitatively compare the performance of four widely used nonlinear methods: time reversibility, sample entropy, delay vector variance, and Lyapunov exponents.
• To evaluate the sensitivity of these methods to nonlinearity levels and their robustness against added noise.
• To compare the effectiveness of nonlinear methods against traditional linear frequency-based methods for distinguishing physiological states (pregnancy vs. labor contractions) in uterine EMG signals.

Main Methods:

• Applied four nonlinear methods (time reversibility, sample entropy, delay vector variance, Lyapunov exponents) to a Henon nonlinear synthetic model with varying complexity.
• Evaluated method performance using sensitivity to complexity and mean square error (MSE) of Monte Carlo instances, with and without added noise, using direct signal application and z-score surrogates.
• Compared nonlinear methods' discrimination performance on real uterine EMG signals against linear methods (MPF, PF, MF).

Main Results:

• Significant differences in performance were observed among the evaluated nonlinear methods.
• Nonlinear methods, specifically time reversibility and Lyapunov exponents, demonstrated superior performance compared to linear frequency-based methods in distinguishing uterine EMG signals.
• Direct application of methods to signals generally yielded better results than using the z-score, with sample entropy being an exception.

Conclusions:

• Nonlinear time series analysis offers a powerful approach for detecting complexity and improving the discrimination of physiological states.
• Time reversibility and Lyapunov exponents are promising nonlinear measures for analyzing complex signals like uterine EMG.
• Further research is needed to fully understand the utility of surrogate data analysis for quantifying nonlinearity levels.