Prediction of autism spectrum disorder diagnosis using nonlinear measures of language-related EEG at 6 and 12 months

Abstract

Early identification of autism spectrum disorder (ASD) provides an opportunity for early intervention and improved outcomes. Use of electroencephalography (EEG) in infants has shown promise in predicting later ASD diagnoses and in identifying neural mechanisms underlying the disorder. Given the high co-morbidity with language impairment in ASD, we and others have speculated that infants who are later diagnosed with ASD have altered language learning, including phoneme discrimination. Phoneme learning occurs rapidly within the first postnatal year, so altered neural substrates either during or after the first year may serve as early, accurate indicators of later autism diagnosis. Using longitudinal EEG data collected during a passive phoneme task in infants with high familial risk for ASD, we compared predictive accuracy at 6-months (during phoneme learning) versus 12-months (after phoneme learning). Samples at both ages were matched in size and diagnoses (n=14 with later ASD; n= 40 without ASD). Using Pearson correlation feature selection and support vector machine with radial basis function classifier, 100% predictive diagnostic accuracy was observed at both ages. However, predictive features selected at the two ages differed and came from different scalp locations. We also report that performance across multiple machine learning algorithms was highly variable and declined when the 12-month sample size and behavioral heterogeneity was increased. These results demonstrate that speech processing EEG measures can facilitate earlier identification of ASD but emphasize the need for age-specific predictive models with large sample sizes in order to develop clinically relevant classification algorithms.