Prediction of group membership in developmental dyslexia, attention deficit hyperactivity disorder, and normal controls using brain morphometric analysis of magnetic resonance imaging

• Published in: Archives of clinical neuropsychology 1996, Vol.11 (6), p.521-528
• Main Authors: Semrud-Clikeman, Margaret, Hooper, Stephen R., Hynd, George W., Hern, Kelly, Presley, Rodney, Watson, Tom
• Format: Article
• Language: English
• ISSN: 0887-6177; 1873-5843
• DOI: 10.1016/0887-6177(95)00044-5

This study explored the utility of using selected brain morphometric indices for predicting group membership for children with developmental dyslexia (n = 10), attention deficit hyperactivity disorder: combined type (n = 10), and a control group (n = 10). Subjects ranged in age from 6.1 to 16 years (M = 10.5 years, SD = 2.8). None of the subjects were diagnosed with mental retardation, nor did any of the subjects have a history of seizure disorder, head trauma, or other neurodevelopmental disorders. WISC-R Full Scale IQ ranged from 87 to 149 (M = 114.4, SD = 13.3) with no significant differences noted between the clinical groups. Six brain regions, as defined by MRI scans, were selected a priori for inclusion in a discriminant function analysis. Reliability of the morphometric measures ranged from 0.94 to 0.97. One significant discriminant function was generated which accounted for about 61.4% of the variance between groups. The predictive discriminant analysis using the six morphometric MRI measurements classified subjects with an overall 60% accuracy with the best accuracy found for the developmental dyslexia and control groups. A predictive discriminant analysis incorporating these six morphometric measures as well as chronological age and FSIQ increased the overall classification accuracy to 87% with the misclassfied subjects assigned to one of the clinical groups. The findings support the presumed neurological basis for many neurodevelopmental disorders. They also underline the importance of including brain morphometric measures in predictive models.