Prediction of Individual Autism Diagnostic Observation Schedule (ADOS) scores based on neural responses during live eye-to-eye contact

Social difficulties are impactful in autism spectrum disorder (ASD), and links between these difficulties and underlying neural processes are active research questions. We present a multivariate classification method for neural data acquired from 36 participants during a live eye-to-eye contact task. Participants were either typically developed (TD) or diagnosed as ASD through gold-standard Autism Diagnostic Observation Schedule (ADOS) evaluation. We hypothesized multivariate classification could discriminate TD vs. ASD based on neural responses. Support vector machine (SVM) classification was able to discriminate between groups during the eye-contact interaction. In addition, it was found that underlying neural patterns contributing to binary classification also predicted measured ADOS scores with high correlation even though ADOS scores were not used for training. The correlation between observed and predicted ADOS scores was 0.72 (p < 0.002) for eye-to-eye contact. These findings suggest neural responses to live eye-to-eye contact are predictive of social symptomatology in ASD.

Methods
Functional near-infrared spectroscopic brain imaging data were collected from a Shimadzu LABNIRS during a face to face interaction task from participants that were categorized as either TD or ASD by a trained clinician. 

Univaraite and Multivariate analyses were used to determine classification of participants as TD or ASD using only FNIRS responses.

Typical GLM methods were used for univariate classification.

For multivariate classification, an SVM classifier was used to predict ADOS values for the ASD group and categorize all participants as either TD or ASD.