High-Dimensional Pattern Classification of Brain Morphometric and DTI Data of Adult ADHD

Abstract

Background: Despite its high prevalence, adult Attention-Deficit/Hyperactivity Disorder (ADHD) has been sparsely investigated with neuroimaging methods. The application of high-dimensional pattern classification methods appears as a promising auxiliary approach to aid in the clinical diagnosis of psychiatric disorders, but such strategy has not yet been applied to the investigation of adult ADHD. Methods: Twenty-two adult, never-treated ADHD patients and 19 age- and gender-matched healthy controls underwent T1-MPRAGE and 64-direction diffusion tensor imaging (DTI) acquisitions using 1.5T MRI scanner. Volumetric maps of gray matter, white matter and ventricular compartments were generated through a robust routine of deformation-based morphometry, whereas a new DTI analysis approach for spatial normalization of tensor fields was used to generate fractional anisotropy (FA) and mean diffusivity (MD) maps. A multivariate classification method based on support vector machine was employed to identify the best set of morphological features that discriminate ADHD patients from controls. Diagnostic measures were obtained through ROC analyses. Results: The best discrimination performance of the classifier was obtained when the FA map was employed in isolation (area under the curve, AUC=0.61). The regions primarily used in the FA classifier were the orbitofrontal cortex and cerebellum. These brain areas were also those with larger differences when voxelwise between-group comparisons of FA data were performed (p<0.001, uncorrected for multiple comparisons). Conclusions: These preliminary results suggest that FA measures are the best indices of brain dysfunction in pattern classification investigations of adult ADHD using MRI. This reinforces the notion of brain connectivity abnormalities underlying the symptoms of ADHD.