Autism Spectrum Disorder Diagnosis framework using Diffusion Tensor Imaging

Abstract

Autism is a complex neurological disorder which affects behavioral and communication skills. Numerous studies were presented suggesting abnormal development of neural networks in the brain in shape, functionality, and/or connectivity. While conventional diagnosis of autism is subjective and requires long time before confirmation, neuro-imaging techniques provide a promising alternative. This paper introduces an automated autism computer-aided diagnosis system based on the connectivity information of the WM tracts. In this CAD system, two consecutive levels of analysis are implemented: Local analysis utilizing diffusion tensor imaging (DTI) data, then getting global decision. Johns Hopkins WM areas' atlas is employed for DTI-volumes segmentation. Correlations of DTI-derived features between different areas in the brain, demonstrating linkage between WM areas were exploited. Then, feature selection extracting the most prominent features among those associations are made. Lastly, an SVM classifier is exploited to produce the final diagnostic decision. We tested our proposed system on a large data set of 263 subjects from NDAR database (141 typically developed subjects: 66 males, and 75 females, and 122 autistics: 66 males, and 56 females), with ages ranging from 96 to 215 months, achieving an overall accuracy of 71%, with Leave-one-subject-out validation.