Objectives: The aims of this study was to investigate the diagnostic value of grey matter anatomical images in adults with ASD.
Methods: Support Vector Machine algorithms (SVM) were used to automatically classify structrual MRI scan in a sample of 54 male adults; 27 with a diagnosis of ASD made using gold standard research interviews and 27 healthy matched controls. The performance of the classification was evaluated using the leave-two-out cross validation approach. To identify the degree to which the classification is driven by autistic symptoms, the test marging was correlated the level of symptom severity.
Results: SVM correctly classified individuals with autism based on their grey matter anatomy with 67% specificity and sensitivity. The test margin (i.e. distance from separating hyperplane) was positively correlated with the level of symptom severity as measured by the Autism Spectrum Quotient (AQ) and Autism Diagnostic Observation Schedule (ADOS) The most discriminating regions between groups were the (1) the limbic system, (2) the fronto-striatal system, and (3) fronto-temporal and fronto-parietal networks. In addition, we found increased regional volumes in components of the cerebellar circuitry.
Conclusions: The brain regions identified by SVM are in agreement with many previous studies employing voxel-based analyses, and have been functionally linked to autistic core symptoms in the past. In addition we have shown that the classification was driven by autistic symptoms rather than autism-unrelated effects. Therefore we propose that the application of SVM on grey matter anatomical scans might provide a valid neuropathological screening tool for ASD to guide and complement a traditional behaviourally guided diagnosis in the future.