A Critical Comparison of Pipelines for Structural Brain Network Analysis

Area of Science:

• Neuroscience
• Network Science
• Computational Biology

Background:

• The human brain functions as a complex network of interconnected regions.
• Structural networks can be derived from cortical thickness (CT) correlations.
• Analyzing these networks requires pipelines involving multiple analytical steps.

Purpose of the Study:

• To critically compare 96 different analysis pipelines for structural brain networks.
• To determine the influence of various pipeline components on network analysis results.
• To identify the most critical steps in transforming CT data into network clusters.

Main Methods:

• Generation of structural networks from cortical thickness (CT) correlations.
• Systematic comparison of 96 distinct analysis pipelines.
• Evaluation of the impact of CT correlation methods and cluster detection algorithms.

Main Results:

• The choice of CT correlation and correction procedures significantly alters results, more so than using absolute vs. all CT correlations.
• Different cluster detection algorithms show substantial variability in their outcomes, with correlations ranging from 0.98 to -0.20 on the same data.
• Pipeline component selection critically influences the final network organization and cluster detection.

Conclusions:

• The selection of specific analysis pipeline components, particularly CT correlation methods and cluster detectors, profoundly impacts structural brain network findings.
• A consensus on optimal pipeline steps is lacking, necessitating the use of multiple, complementary methods.
• Further research is required to establish theory-driven recommendations for robust neuroscientific network analysis.