A Novel Method for Multi-subject fMRI Data Analysis: Independent Component Analysis with Clustering Embedded (ICA-CE)

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Data Science

Background:

• Accurate analysis of multi-subject functional magnetic resonance imaging (fMRI) data is crucial for understanding brain functional networks (FNs).
• Traditional independent component analysis (ICA) methods struggle with extracting individual FNs when subject classes are unknown.
• Existing methods like clusterwise ICA (C-ICA) have limitations in clustering performance with complex datasets.

Purpose of the Study:

• To develop a novel method, independent component analysis with clustering embedded (ICA-CE), for unsupervised or semi-supervised estimation of individual FNs and subject clustering.
• To improve the accuracy of brain FNs extraction and subject classification from fMRI data.

Main Methods:

• Proposed ICA-CE method integrates subject clustering and ICA for simultaneous estimation of individual FNs and group-level clustering.
• Validated using simulated fMRI data with varying properties.
• Applied to task-related fMRI data from the Human Connectome Project (HCP).

Main Results:

• ICA-CE demonstrated superior clustering performance compared to group ICA with K-means and C-ICA on simulated data.
• Achieved over 90% mean accuracy in extracting individual FNs from simulated data.
• Showcased higher clustering accuracy and identified task-related class-specific FNs on HCP data.

Conclusions:

• ICA-CE is an effective method for accurate brain FNs estimation and multi-subject clustering.
• The approach holds promise for clinical applications, aiding in the analysis of brain disorders and subject stratification.