Extended nonnegative matrix factorization for dynamic functional connectivity analysis of fMRI data

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Dynamics Analysis

Background:

• Dynamic functional connectivity (DFC) analysis using functional magnetic resonance imaging (fMRI) is crucial for understanding brain dynamics.
• Traditional nonnegative matrix factorization (NMF) has limitations in DFC analysis due to its nonnegative constraint on input data.

Purpose of the Study:

• To introduce an extended NMF (eNMF) method that accommodates negative values in input and decomposed matrices for DFC analysis.
• To evaluate the performance of eNMF in analyzing simulated and real resting-state fMRI data compared to K-means.

Main Methods:

• Development and application of the extended nonnegative matrix factorization (eNMF) algorithm.
• Analysis of simulated and real resting-state fMRI data.
• Comparison of eNMF with K-means clustering for DFC analysis.

Main Results:

• eNMF successfully decomposed mixed-sign matrices into positive and mixed-sign components in simulated data.
• eNMF extracted more accurate brain state patterns and estimated superior DFC temporal properties than K-means.
• Real fMRI data analysis showed eNMF provided richer temporal DFC measures and greater sensitivity to intergroup differences.

Conclusions:

• The proposed eNMF method offers significant improvements for DFC analysis, overcoming limitations of traditional NMF.
• eNMF is more effective than K-means in identifying brain states and their temporal dynamics from fMRI data.
• Preliminary findings suggest potential sex-based differences in DFC, with females exhibiting altered relaxation and cognitive process patterns.