Alcohol use effects on adolescent brain development revealed by simultaneously removing confounding factors,...

Area of Science:

• Neuroscience
• Machine Learning
• Neuroimaging

Background:

• Generalized Additive Models (GAM) are common for analyzing brain MRI data, removing confounding factors before cohort analysis.
• The National Consortium on Alcohol and NeuroDevelopment in Adolescence (NCANDA) has used GAM to study alcohol's effects on the adolescent brain.

Purpose of the Study:

• To test the hypothesis that pre-analysis confounding factor removal discards crucial information for distinguishing drinking behaviors.
• To introduce a novel machine learning model that simultaneously trains a GAM and a classifier for enhanced neuromorphometric pattern identification.

Main Methods:

• A machine learning model was developed to jointly train a GAM and a generic classifier.
• The model analyzed macrostructural MRI and microstructural diffusion tensor imaging (DTI) metrics from the NCANDA dataset (N=705).
• Compared the proposed model against traditional group analysis and other machine learning techniques.

Main Results:

• The proposed machine learning approach identified a distinct pattern of eight brain regions associated with alcohol misuse in adolescents.
• Classification accuracy for high-drinking adolescents was superior using the identified pattern compared to regions from alternative methods.
• The joint model approach proved impartial to confounding factors, relevant to drinking behaviors, and consistent with existing alcohol literature.

Conclusions:

• Simultaneously training GAM and classifiers offers an improved method for identifying neuroimaging patterns related to alcohol misuse in adolescents.
• This integrated approach enhances the accuracy of distinguishing between adolescent drinking behaviors using neuroimaging data.
• The findings support the utility of machine learning in uncovering subtle neuromorphometric differences linked to substance use.