Robust features for the automatic identification of autism spectrum disorder in children

Background: It is commonly reported that children with autism spectrum disorder (ASD) exhibit hyper-reactivity or hypo-reactivity to sensory stimuli. Electroencephalography (EEG) is commonly used to study neural sensory reactivity, suggesting that statistical analysis of EEG recordings is a potential means of automatic classification of the disorder. EEG recordings taken from children, however, are frequently contaminated with large amounts of noise, making analysis difficult. In this paper, we present a method for the automatic extraction of noise-robust EEG features, which serve to quantify neural sensory reactivity. We show the efficacy of a system for the classification of ASD using these features. Methods: An oddball paradigm was used to elicit event-related potentials from a group of 19 ASD children and 30 typically developing children. EEG recordings were taken and robust features were extracted. A support vector machine, logistic regression, and a naive Bayes classifier were used to classify the children as having ASD or being typically developing. Results: A classification accuracy of 79% was achieved, making our method competitive with other automatic diagnosis methods based on EEG. Additionally, we found that classification performance is reduced if eye blink artifacts are removed during preprocessing. Conclusions: This study shows that robust EEG features that quantify neural sensory reactivity are useful for the classification of ASD. We showed that noise-robust features are crucial for our analysis, and observe that traditional preprocessing methods may lead to poor classification performance in the face of a large amount of noise. Further exploration of alternative preprocessing methods is warranted.