An Orthogonal Semi-Nonnegative Matrix Factorization Method for Dynamic Functional Connectivity Analysis and Its...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Data Analysis

Background:

• Dynamic functional connectivity (dFC) analysis tracks changing brain region interactions over time.
• Non-negative matrix factorization (NMF) is useful but limited to positive values, excluding anti-correlations.
• Existing methods cannot fully capture the complexity of brain functional correlations.

Purpose of the Study:

• To develop an advanced NMF method for analyzing mixed-sign dFC data, overcoming limitations of existing approaches.
• To enhance the uniqueness and interpretability of dynamic functional connectivity states.
• To identify reproducible dFC biomarkers for neurological disorders, specifically schizophrenia.

Main Methods:

• Proposed an orthogonal semi-nonnegative matrix factorization (OSemiNMF) method to process mixed-sign dFC data.
• Incorporated an orthogonality constraint on dFC states (bases) to improve their distinctiveness.
• Validated the method on simulated datasets and real resting-state fMRI data from healthy controls and schizophrenia patients.

Main Results:

• OSemiNMF outperformed comparison methods in capturing dFC states and transitions on simulated data.
• Identified reproducible dFC states and transitions across multiple resting-state fMRI datasets.
• Schizophrenia patients exhibited less time in high-connectivity states compared to healthy controls, with sub-cortical connectivity being a key differentiator.

Conclusions:

• The OSemiNMF method provides a robust framework for analyzing complex brain dynamics in functional connectivity data.
• This approach successfully identified reproducible dFC biomarkers associated with schizophrenia.
• The findings highlight the potential of OSemiNMF for advancing our understanding of brain disorders.