A unified framework for group independent component analysis for multi-subject fMRI data.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Analysis

Background:

• Independent Component Analysis (ICA) is increasingly used for functional magnetic resonance imaging (fMRI) data analysis.
• Extending ICA to group inferences is challenging, with current models like GIFT and tensor PICA making different assumptions and potentially yielding divergent results.
• There is a lack of methods to assess and select appropriate group ICA model structures for real fMRI data.

Purpose of the Study:

• To propose a unified framework for estimating and comparing group ICA models with varying spatio-temporal structures.
• To develop a method for selecting the most appropriate group ICA model for fMRI data.
• To enable robust group-level fMRI data analysis and comparison.

Main Methods:

• A unified framework accommodating diverse group structures, including GIFT and tensor PICA as special cases.
• Maximum Likelihood (ML) estimation with a modified Expectation-Maximization (EM) algorithm.
• Likelihood Ratio Tests (LRT) for model comparison, selection, goodness-of-fit assessment, and testing group differences.

Main Results:

• The proposed framework allows for the estimation and comparison of various group ICA models.
• Likelihood Ratio Tests effectively compare models and assess goodness-of-fit for fMRI data.
• Simulation studies validated the method's performance across different group spatio-temporal structures.

Conclusions:

• The developed framework provides a robust approach for group ICA in fMRI.
• The LRT enables rigorous model selection and group comparison in neuroimaging studies.
• This method enhances the analysis of complex neural processing, as demonstrated in a Zen meditation fMRI study.