Structural differences in adolescent brains can predict alcohol misuse.

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Adolescent alcohol misuse (AAM) is linked to disrupted brain development.
• Predicting AAM from neuroimaging data is challenging due to small effect sizes.

Purpose of the Study:

• To predict adolescent alcohol misuse (AAM) phenotypes using brain structural MRI data.
• To identify brain regions and features predictive of AAM.
• To evaluate the performance of different machine learning models for AAM prediction.

Main Methods:

• Longitudinal machine learning study using the IMAGEN dataset (n≈1182).
• Utilized T1-weighted imaging and Diffusion Tensor Imaging (DTI) data.
• Predicted 10 AAM phenotypes at age 22 using brain MRI features from ages 14, 19, and 22.

Main Results:

• Machine learning models predicted AAM with 73-78% accuracy.
• Binge drinking was the most predictable AAM phenotype.
• Predictive brain features were identified in CSF, white matter tracts (corpus callosum, internal capsule, brain stem), and cortical/cingulate regions.

Conclusions:

• Structural brain differences can predict AAM and may precede its onset.
• Machine learning models, particularly SVM and Gradient Boosting, show promise in predicting AAM.
• Phenotype choice, ML model, and confound correction are critical for AAM studies.