Predictive models for autism spectrum disorder based on multiple cortical features

Autism spectrum disorder (ASD) is a neurodevelopmental disorder with a wide phenotypic range, often affecting personality and communication. Previous surface-based morphometry (SBM) work only used cortical thickness to generate classification model between ASD and normal control. The goals of this study were twofold: 1) to construct predictive models for ASD, based on combination any of these five cortical measurements (cortical thicknesses (T), mean curvature (MC), Gaussian curvature (GC), folding index (FI), and curvature index(CI)) extracted from SBM, 2) and to compare these models. Our study included 22 subjects with ASD (mean age 9.2±2.1 years) and 16 volunteer controls (mean age 10.0±1.9 years). Using SBM, we obtained T, MC, GC, FI and CI for 66 brain structures for each subject. Then, we combined any of these five cortical measurements and obtained 31 combinations as classification inputs. To generate predictive models, we employed three machine-learning techniques: support vector machines (SVMs), functional trees (FTs), and logistic model trees (LMTs). We found that “thickness + mean curvature”-based classification model was modest superior to that based on thickness-based features when LMT employed, and curvature only provide limited information to thickness in ASD predictive model. Our result suggested that ASD may be cortical thickness abnormal disorder rather than cortical curvature abnormal disorder.