Joint, Partially-Joint, and Individual Independent Component Analysis in Multi-Subject fMRI Data

Area of Science:

• Biomedical Engineering
• Neuroscience
• Data Analysis

Background:

• Joint analysis of multi-subject brain imaging data is crucial for understanding complex neurological conditions.
• Existing methods struggle to differentiate between sources shared across all subjects, subsets, or individual subjects.

Purpose of the Study:

• To develop a novel source extraction method for joint/partially-joint/individual multiple datasets unidimensional (JpJI-MDU) data.
• To introduce a feature for characterizing source jointness and type.

Main Methods:

• A deflation-based algorithm using higher-order cumulants to analyze the JpJI-MDU model.
• Maximization of a cost function leading to an eigenvalue problem solved via thin-SVD.
• Introduction of the JpJI-feature for source identification and classification.

Main Results:

• The algorithm achieved 95% and 100% accuracy in classifying sources in 2-class and 3-class simulations, respectively.
• Meaningful joint and partially-joint sources were successfully extracted from real fMRI datasets.
• Extracted sources align with existing neuroscience findings.

Conclusions:

• The JpJI-MDU source model accurately describes multi-subject brain imaging datasets.
• The proposed algorithm enhances the accuracy of source extraction for multi-subject data.
• The algorithm is robust and user-friendly, avoiding difficult parameter tuning.