A COMPARISON OF NETWORK DEFINITIONS FOR DETECTING SEX DIFFERENCES IN BRAIN CONNECTIVITY USING SUPPORT VECTOR MACHINES

Area of Science:

• Neuroscience
• Computational Biology
• Brain Imaging

Background:

• Human brain connectomics investigates neural connections using diverse methodologies.
• Previous research has identified sex differences within the structural connectome.
• The precise definition of brain regions and connections influences connectome reconstruction.

Purpose of the Study:

• To assess how variations in defining regions of interest, connection parameters, and normalization methods affect structural connectome calculations.
• To determine the impact of these reconstruction choices on classifying subjects by sex using machine learning.

Main Methods:

• Evaluation of different region definitions (atlas-based vs. cluster-based).
• Analysis of connection definitions and normalization strategies (total connectivity, region size).
• Application of a support vector machine (SVM) classifier to differentiate between sexes based on reconstructed connectomes.

Main Results:

• Cluster-based regions yielded higher sex classification accuracy compared to atlas-based regions.
• Finer cortical partitions improved classification performance.
• Dilating regions of interest before network computation enhanced sex classification accuracy.

Conclusions:

• The methodology for defining regions of interest is critical for accurate structural connectome analysis.
• Specific choices in connectome reconstruction, particularly using cluster-based and finer partitions, enhance the detection of sex differences.
• These findings have implications for understanding sex-specific brain organization and improving diagnostic tools.