Date of Award


Document Type


Degree Name

Master of Science (MS)



First Advisor

Alexandra S. Potter

Second Advisor

Sayamwong Hammack


Psychostimulant medication is the first line of treatment for Attention deficit/hyperactivity disorder (ADHD). Despite the prevalence of ADHD, there is a lack of understanding of the underlying neurophysiological mechanisms of the disorder and its pharmacological treatments. Existing neuroimaging research shows some consistent structural differences in ADHD, but it can be difficult to discern what is relevant. Machine learning algorithms present a novel way of analyzing a large amount of data by making predictions based on pattern detection.

The present study applied an elastic-net logistic machine learning model to structural magnetic resonance imaging (sMRI) data from the Adolescent Brain Cognitive Development (ABCD) Study to predict ADHD diagnosis and psychostimulant medication use. Based on Matthews correlation coefficient and receiver operating characteristic curves, the model achieved modest success in classifying ADHD diagnosis and psychostimulant use amongst the whole sample regardless of the inclusion of covariates, along with psychostimulant use amongst males assigned at birth and ADHD-negative subjects. Classifying psychostimulant use was consistently more successful than classifying ADHD diagnosis. In line with existing ADHD research, important features for prediction included subregions in the frontal, temporal, and parietal lobes. Correlations between several subregional volumes and stimulant use had opposing directions in the ADHD-positive sample compared with other groups, implying an ADHD-dependent effect of medication. The finding that stimulant use is more detectable from sMRI data than ADHD urges further investigation of these commonly prescribed drugs and their relationship to the brain, especially in children and adolescents.



Number of Pages

50 p.