How to Improve the Reliability of Aperiodic Parameter Estimates in M/EEG: A Method Comparison

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Broadband aperiodic brain activity (1/f phenomenon) analysis is increasingly important.
• Current methods like the FOOOF toolbox require separating periodic and aperiodic activity, with peak detection sensitivity being a concern.

Purpose of the Study:

• Investigate the impact of analytic choices, such as the number of peaks, on aperiodic parameter estimation reliability.
• Propose and validate a novel method, censored regression, to improve the robustness of aperiodic parameter estimates.

Main Methods:

• Evaluated the effects of varying peak detection parameters in the FOOOF toolbox on aperiodic parameter reliability (intercept, slope).
• Developed and applied a censored regression approach, removing a theory-driven frequency range before parameter estimation.
• Assessed methods using internal consistency, outlier generation, and effect detection across resting-state and task-based datasets.

Main Results:

• Increasing the number of detected peaks in FOOOF decreased the reliability of aperiodic intercept and slope estimates.
• The proposed censored regression method demonstrated more reliable and robust aperiodic parameter estimates compared to FOOOF.
• Censored regression effectively avoided overfitting issues observed with increased peak flexibility.

Conclusions:

• Analytic choices in spectral parametrization significantly impact the reliability of broadband aperiodic brain activity metrics.
• Censored regression offers a more robust and reliable alternative for estimating aperiodic brain activity parameters, enhancing M/EEG data analysis.
• This modified approach improves the accuracy and consistency of findings derived from aperiodic brain activity.