Identifying brain pathological abnormalities of autism for classification using diffusion tensor imaging

Autism is a neurodevelopmental disorder that is characterized by impairments in behavioral and communication skills. Typically, diagnosis is made in early childhood on the basis of physician visits and interviews and observation of set of developmental behaviors. Although there are many studies that identified different aspects of brain abnormalities, rather in shape, functionality, and/or connectivity, still it was difficult to find commonalities across all. The processes of providing diagnosis using conventional means is hard and lengthy, as it requires long observations and repetitive assessments of the autism spectrum disorder (ASD) subject; thus neuroimaging techniques offer a fast and objective alternative. In this work, we utilize brain imaging data to analyze the white matter (WM) connectivity information and classify autistic subjects using automated CAD framework. Diffusion tensor imaging (DTI) connectivity features are extracted, segmented into various WM brain areas set by Johns Hopkins University WM atlas, reduced, and a feature selection was used to find the most distinctive features that will be used in the final classification step. Our proposed framework was applied and tested on a large dataset of 225 subjects obtained from ABIDE-II database, from 5 different sites (125 ASDs: 104 males and 21 females, and 100 typically developed: 76 males and 24 females), with ages between 5.128 and 46.6 years, obtaining an accuracy of 70%, with leave-one-out cross-validation technique. These promising results highlight the ability to identify brain connectivity abnormalities using DTI data.