Model selection for spectral parameterization

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Neurophysiological brain activity has both rhythmic and arrhythmic components.
• Current spectral analysis methods for neural recordings lack robustness and reproducibility due to user-dependent parameter selection.

Purpose of the Study:

• To develop a principled, data-driven approach for spectral parameterization of neurophysiological data.
• To enhance the reliability and interpretability of findings from neural recordings.

Main Methods:

• Introduced a model selection approach using the Bayesian Information Criterion for static and time-resolved spectral parameterization.
• Validated the method with ground-truth and empirical magnetoencephalography (MEG) recordings.

Main Results:

• The data-driven model selection significantly improved the specificity and sensitivity of spectral and spectrogram decompositions.
• The approach demonstrated effectiveness even in non-stationary neural data.
• Reduced reliance on user expertise and subjective parameter choices.

Conclusions:

• The proposed spectral decomposition with data-driven model selection offers a more robust and reproducible method for analyzing neurophysiological data.
• This approach facilitates more interpretable research findings in neuroscience.