Automatic Detection of Autism Spectrum Disorder by Tracing the Disorder Co-morbidities

Preliminary behavioral studies have investigated the co-morbidities in Autism Spectrum Disorder (ASD) for behavioral identification of individuals. However, the studies based on automatic detection of ASD affected individuals by tracing the co-morbidity are few in number. The present study examined the risk-perception (one of the co-morbid features) and computed the neurological performance, using Event Related Potential (ERP), in risky situations for automatic identification of the disorder. The motive of the present study is: (i) to provide insights into the mechanism underlying risk-perception in ASD, and (ii) to automatically detect ASD individuals on the basis of their risk-perception. The ERP features such as latency, amplitude, and sum of signed differences have been extracted and analyzed statistically as well as using machine learning based Support Vector Machine (SVM) classifier. The statistical results depicted that both ASD and Typically Developing (TD) individuals showed larger P300 response to the risky stimuli [F=3.423, p=0.031; F=1.306, p=0.001]. However, latencies were found in response patterns of ASD. The classification results showed that the combination of different features provided high accuracy (99%) in neurologically differentiating ASD from Typically Developing (TD). The present work findings reveal that risk knowledge is persistent, but delayed in ASD. This co-morbid trait has helped in automatic identification of ASD.