Data-Driven and Predefined ROI-Based Quantification of Long-Term Resting-State fMRI Reproducibility.

Area of Science:

• Neuroimaging
• Neuroscience
• Brain Connectivity

Background:

• Resting-state functional magnetic resonance imaging (fMRI) is crucial for neuroscience and clinical research.
• Significant variability exists in resting-state functional connectivity strength and spatial extent across sessions.
• Reproducibility of resting-state fMRI is often evaluated using predefined regions-of-interest (ROIs), potentially biasing results.

Purpose of the Study:

• To compare the reproducibility of resting-state fMRI using data-driven versus predefined ROI-based quantification.
• To investigate the impact of ROI definition on reproducibility measures.
• To determine if data-driven methods offer a more accurate assessment of resting-state fMRI reproducibility.

Main Methods:

• Employed a support vector machine (SVM)-based technique to identify functionally connected voxels for data-driven reproducibility analysis.
• Quantified reproducibility using all voxels within predefined ROIs for comparison.
• Analyzed within-subject and between-subject reproducibility.

Main Results:

• Data-driven analysis using SVM-identified voxels yielded moderate to substantial within-subject and reasonable between-subject reproducibility.
• Increasing ROI size in predefined ROI analysis did not consistently improve reproducibility.
• Reproducibility was generally higher when using identified functionally connected voxels compared to all voxels in typical ROIs.

Conclusions:

• Data-driven methods, particularly using SVM-identified functionally connected voxels, provide a more robust measure of resting-state fMRI reproducibility.
• Conventional ROI-based analyses may underestimate the true reproducibility of resting-state fMRI.
• Findings highlight the importance of considering analysis methodology in interpreting resting-state fMRI reproducibility.