A statistical framework for neuroimaging data analysis based on mutual information estimated via a gaussian copula.

Area of Science:

• Neuroimaging
• Information Theory
• Computational Neuroscience

Background:

• Information theory offers a powerful framework for neuroimaging data analysis.
• Estimating information-theoretic quantities in practice is a significant challenge in neuroimaging.
• Existing methods lack flexibility in handling diverse data types and modalities.

Purpose of the Study:

• To develop a novel, robust, and computationally efficient statistical framework for estimating information-theoretic quantities in neuroimaging.
• To enable a unified treatment of discrete, continuous, unidimensional, and multidimensional variables.
• To facilitate direct comparisons of brain and behavioral responses across different recording modalities.

Main Methods:

• Combined the statistical theory of copulas with closed-form solutions for Gaussian variable entropy.
• Developed a multivariate statistical framework for effect size estimation on a common scale.
• Validated the estimation technique as a statistical test for discrete and continuous stimulus features in neuroimaging.

Main Results:

• The novel method provides a general, flexible, and robust framework for neuroimaging data analysis.
• Demonstrated the ability to quantify modulations of amplitude and direction for vector quantities.
• Showcased the measurement of emergent information over time in evoked responses using M/EEG signals.
• Successfully applied multivariate analyses to MEG and EEG data, including temporal interactions.

Conclusions:

• The developed framework significantly advances the practical application of information theory in neuroimaging.
• The method allows for a more comprehensive and unified analysis of brain and behavioral data.
• Open-source code is provided to facilitate the adoption and further development of these techniques.