Experimental Comparisons of Sparse Dictionary Learning and Independent Component Analysis for Brain Network Inference...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Data Analysis

Background:

• Functional Magnetic Resonance Imaging (fMRI) is crucial for mapping brain activity.
• Inferring functional brain networks from fMRI data is complex due to overlapping signals.
• Independent Component Analysis (ICA) and Sparse Dictionary Learning (SDL) are common methods for this task.

Purpose of the Study:

• To comprehensively compare the performance of different ICA and SDL variants.
• To evaluate their efficacy in inferring functional brain networks from synthesized fMRI data with ground-truth.
• To provide guidelines for selecting and interpreting network analyses in fMRI connectomics.

Main Methods:

• Utilized four variants of Independent Component Analysis (ICA).
• Employed three variants of Sparse Dictionary Learning (SDL) algorithms.
• Conducted subject-level comparisons using synthesized fMRI data with known ground-truth.

Main Results:

• ICA methods performed well with minor spatial overlaps but struggled with moderate to severe overlaps.
• SDL algorithms demonstrated consistent performance across all levels of spatial overlap.
• SDL methods significantly outperformed ICA methods when functional networks exhibited moderate or severe spatial overlaps.

Conclusions:

• SDL methods offer superior robustness and accuracy compared to ICA for fMRI network inference, especially with overlapping networks.
• This study enhances the understanding of ICA and SDL algorithms in fMRI data analysis.
• Findings provide critical guidelines for constructing and interpreting functional brain networks in fMRI-based connectomics.