Attention-deficiyhyperactivity disorder (ADHD) has become a common functional disorder in children and adolescents. However, diagnosing this condition is challenging since it relies exclusively in subjective biased clinical evaluation. Despite ongoing research, findings of structural changes in specific brain areas are inconsistent and hardly reproducible. This study explores how local and regional relationships may provide an alternative description of ADHD by a curvelet analysis of a set of brain regions expressing ADHD differences. Curvelets may capture differences at several scales, either circumvolution variations or local geometrical organization changes. Once data are projected to the curvelet space, each frequency subband is approximated by a generalized Gaussian with parameters mean μ, spread α, and decay β. Several binary classifiers were then trained and independently applied with each of these parameters to distinguish ADHD and typically developing (TD) conditions in 164 subjects from the open ADHD200 database. Classifier performance was evaluated using precision, sensitivity, and specificity under a 70/30 validation scheme. Best classification accuracy was 72% with the spread α parameter from the left putamen applying a Logistic Regression Classifier. As these differences come from high scales, they might be attributed to different sulci morphology at local level.