Neuroanatomical morphometric characterization of sex differences in youth using statistical learning

Area of Science:

• Neuroscience
• Statistical Learning
• Neuroimaging

Background:

• Univariate analysis has limitations in detecting subtle, regional neuroanatomical sex differences.
• Multivariate approaches are necessary to model spatial complexity and feature interactions in the brain.

Purpose of the Study:

• To develop and validate a multivariate statistical learning model for predicting sex from neuroanatomical features.
• To identify specific brain regions and cortical measures that are significant predictors of sex in youth.

Main Methods:

• A support vector machine (SVM) classifier was trained on MRI-derived regional neuroanatomical features from the Philadelphia Neurodevelopmental Cohort (PNC).
• The model was validated on an independent dataset from the multi-site Pediatric Imaging, Neurocognition and Genetics (PING) cohort.
• Cortical thickness, using the Destrieux atlas, was compared with other atlases for sex discrimination.

Main Results:

• The trained model achieved 83% cross-validated accuracy and 77% accuracy on the independent dataset.
• Cortical thickness in the middle occipital lobes and angular gyri were identified as major predictors of sex.
• The Destrieux atlas provided better discrimination between sexes compared to other tested atlases.

Conclusions:

• Multivariate statistical learning effectively identifies subtle neuroanatomical sex differences in youth.
• Interactions among brain features are crucial for characterizing sex differences, surpassing classical regression.
• Cortical thickness, particularly using the Destrieux atlas, is a valuable measure for sex discrimination in neuroimaging studies.