Functional near infrared spectroscopy for brain functional connectivity analysis: A graph theoretic approach

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Functional Near-Infrared Spectroscopy (fNIRS) is an optical neuroimaging technique measuring brain activity via hemodynamic changes.
• fNIRS signals are susceptible to noise and artifacts, posing challenges for accurate brain connectivity analysis.
• Graph theory offers a framework for analyzing brain network geometry and functional parameters.

Purpose of the Study:

• To assess the reproducibility of graph measurements in functional neuroimaging using fNIRS data.
• To investigate the impact of noise removal techniques and correlation methods on brain network reproducibility.
• To determine optimal parameters for reliable individual-level brain connectivity analysis.

Main Methods:

• Examined reproducibility of graph measurements using test-retest variability (TRT) on fNIRS data.
• Compared two noise removal methods (CBSI, TDDR) and two correlation types (Pearson, Cross Correlation).
• Analyzed whole-brain network architectures at densities from 5% to 50%.

Main Results:

• High test-retest variability was observed for global measurements in binary networks, especially at low densities.
• Weighted networks demonstrated significantly better reproducibility for graph measures.
• Correlation type, absolute correlation value, and weight calculation method substantially impacted test-retest values.

Conclusions:

• Normalized global graph measurements are reliable when using weighted networks with absolute cross-correlation.
• Node definition techniques for noise removal were not essential for reproducible normalized graph measures.
• This research provides insights into optimizing fNIRS data analysis for robust brain connectivity assessment.