Deep Learning Prediction and Visualization of Gender Related Brain Changes from Longitudinal Structural MRI Data in...

Area of Science:

• Neuroimaging
• Developmental Neuroscience
• Machine Learning

Background:

• Deep learning models show promise for neuroimaging analysis.
• Previous research primarily focused on adult neuroimaging data.
• The Adolescent Brain and Cognitive Development (ABCD) dataset offers a valuable resource for studying adolescent brain development.

Purpose of the Study:

• To predict gender using structural MRI data from the ABCD dataset.
• To identify age-related gender differences in adolescent brain structure.
• To explore the potential of deep learning in understanding adolescent brain development.

Main Methods:

• Utilized deep learning algorithms on 3D structural MRI data.
• Employed the Adolescent Brain and Cognitive Development (ABCD) dataset.
• Analyzed gender prediction accuracy and its correlation with age.

Main Results:

• Achieved high gender prediction accuracy (>94%), which improved with age.
• Identified frontal and temporal lobe regions as key discriminators for gender prediction.
• Revealed broader visual, cingulate, and insular regions associated with age-specific gender differences.

Conclusions:

• Demonstrated robust, age-related, gender-specific structural brain changes in adolescents.
• Highlighted the potential of deep learning for analyzing neuroimaging data in developmental studies.
• Suggested future research directions linking these brain changes to behavioral and environmental factors.