Robust Nonlinear Causality Analysis of Nonstationary Multivariate Physiological Time Series.

Area of Science:

• Biomedical Signal Processing
• Time Series Analysis
• Neuroscience

Background:

• Quantifying relationships between simultaneously observed time series is crucial for understanding signal interactions.
• Biomedical signals are often nonstationary and susceptible to outliers and artifacts, complicating analysis.

Purpose of the Study:

• To introduce a robust time-varying generalized partial directed coherence (rTV-gPDC) function for analyzing biomedical signals.
• To develop a method capable of revealing directed interactions in potentially nonstationary and artifact-laden time series data.

Main Methods:

• The proposed method utilizes a robust estimator for time-varying autoregressive (TVAR) parameters.
• It quantifies linear relationships using rTV-gPDC and approximates nonlinear causality from TVAR residuals using a piecewise linear time-varying moving-average model.

Main Results:

• Extensive simulations demonstrated the performance of the rTV-gPDC method.
• Application to a neurophysiological study involving intracranial pressure, arterial blood pressure, and PtiO2 measurements showcased its real-world applicability.

Conclusions:

• The rTV-gPDC successfully revealed causal patterns consistent with cardiosudoral mechanisms and offered potential new insights into traumatic brain injuries.
• The method is broadly applicable for uncovering interactions in diverse biomedical signal processing applications.