Pseudo-Bootstrap Network Analysis-an Application in Functional Connectivity Fingerprinting.

Area of Science:

• Neuroscience
• Network Science
• Medical Imaging Analysis

Background:

• Brain parcellation is crucial for network analysis, but template-based methods often fail to align with individual functional brain regions.
• This misalignment leads to inconsistent network analysis results, hindering reliable statistical inferences.
• Existing methods struggle with accurately parcellating the brain into functionally relevant areas for each individual.

Purpose of the Study:

• To develop a novel method for brain parcellation that overcomes the limitations of template-based approaches.
• To improve the reliability and statistical power of brain network analysis, particularly in individual subjects.
• To enhance the accuracy of identifying individuals across different scan sessions using functional connectivity.

Main Methods:

• A pseudo-bootstrap (PBS) parcellation strategy was developed, involving multiple applications of roughly equal-sized random parcellations to individual subjects.
• This method generates a pseudo-bootstrap sample of the brain network for robust statistical inferences.
• The PBS method was applied to the Human Connectome Project resting state dataset to assess its performance in individual identification.

Main Results:

• The variation in mean global network metrics derived from PBS sampling was found to be smaller than inter-subject or within-subject variations from diffusion MRI scans.
• Using mean global network metrics from PBS sampling consistently resulted in higher intra-class correlations compared to using a single random parcellation.
• Individual identification across scan sessions achieved an accuracy rate of approximately 90% based on the mean functional connectivity of PBS samples.

Conclusions:

• The pseudo-bootstrap parcellation method offers a robust approach to brain network analysis, enhancing statistical inference and reliability.
• PBS sampling effectively reduces variability, leading to more stable network metrics.
• This method demonstrates significant potential for applications like individual identification, even with simple network properties such as functional connectivity.