Non-linear auto-regressive models for cross-frequency coupling in neural time series

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Cross-frequency coupling (CFC) is crucial for understanding neural communication.
• Existing methods for detecting CFC in neural time series have limitations, including issues with filtering and the Hilbert transform.
• A need exists for robust, data-driven methods to quantify CFC and its properties.

Purpose of the Study:

• To develop and validate a reliable parametric method for detecting and quantifying cross-frequency coupling (CFC) in neural time series.
• To address limitations of existing CFC detection methods, particularly concerning filtering and wide-band signals.
• To provide a data-driven approach for model selection and parameter comparison in CFC analysis.

Main Methods:

• Utilized non-linear auto-regressive models to create a generative, parametric model of time-varying spectral content.
• Modeled the entire signal spectrum simultaneously to avoid filtering pitfalls and Hilbert transform issues.
• Employed a probabilistic framework providing a likelihood-based goodness-of-fit score for model evaluation.

Main Results:

• The proposed method successfully replicated previous CFC findings across human and rodent neurophysiological datasets.
• New insights were revealed, including the influence of slow oscillation amplitude on CFC.
• Simulations demonstrated the parametric method's ability to detect neural couplings in shorter signals compared to non-parametric approaches.
• Likelihood analysis enabled optimal filtering parameter selection and estimation of coupling directionality.

Conclusions:

• The developed parametric, model-based approach offers a robust and unique solution for CFC detection and quantification.
• This method enhances the analysis of neural oscillations, providing deeper insights into brain dynamics.
• The approach facilitates accurate parameter estimation, including directionality, crucial for understanding neural communication.