Looking through the windows: a study about the dependency of phase-coupling estimates on the data length

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Real-time characterization of functional connectivity (FC) is crucial for brain-computer interfaces and neurofeedback.
• Rapid detection of statistical dependencies in short data windows is needed for dynamic FC analysis.
• Phase-coupling (PC) is a key neural coupling mode for quantifying FC.

Purpose of the Study:

• To evaluate the reliability of FC estimation based on data length using extensive simulations.
• To investigate the impact of various factors on phase-coupling (PC) metric performance.
• To establish guidelines for real-time frequency-specific PC assessment.

Main Methods:

• Generated synthetic data with varied data lengths, signal-to-noise ratios (SNR), and spectral analysis methods (Hilbert, Fourier).
• Compared seven phase-coupling (PC) metrics, including iPLV, PLV, PLI, and coherence variants.
• Analyzed performance across different fractional bandwidths and phase difference values.

Main Results:

• A data window of 5-8 cycles (e.g., 500-800 ms at 10 Hz) with SNR ≥ 10 dB is generally needed for reliable PC estimates.
• Hilbert-based spectral analysis approaches demonstrated higher performance compared to Fourier-based methods.
• Narrow frequency band analysis necessitates longer data windows for accurate results.

Conclusions:

• Findings provide a basis for developing best-practice guidelines for real-time PC assessment.
• Optimizing data window length is critical for accurate FC state dynamics in real-time applications.
• The study highlights the importance of choosing appropriate spectral analysis techniques for reliable neural coupling measures.