EEG-Driven Stratification of Autism Spectrum Disorder with and without Epilepsy

Oral Presentation
Friday, May 11, 2018: 10:55 AM
Grote Zaal (de Doelen ICC Rotterdam)
H. Bruining1, S. Simpraga2, E. Juarez2, J. J. Sprengers3, S. S. Poil2, H. Mansvelder2 and K. Linkenkaer-Hansen4, (1)Brain Centre Rudolf Magnus, Amsterdam, Netherlands, (2)CNCR, VU Medical Center, Amsterdam, Amsterdam, Netherlands, (3)Brain Center Rudolf Magnus, Department of Psychiatry, UMC Utrecht, Utrecht, Netherlands, (4)Department of Integrative Neurophysiology, CNCR, VU University Amsterdam, Amsterdam, Netherlands

Clinical heterogeneity makes understanding of autism spectrum disorder (ASD) challenging and complicates development of successful treatment. Neurophysiologically, excitatory-inhibitory (E/I) imbalance and disturbed neuronal oscillations have been proposed as central mechanisms behind ASD.


We aimed to use EEG biomarkers to assess neural disturbances and make ASD triage more precise than is currently feasible using current behavioral and sensory scales. We further investigated whether ASD children with (ASD-EPI) or without epilepsy (ASD) can be stratified using integrative indices derived from comprehensive EEG biomarker mapping and machine-learning.


90 ASD and 30 typically developing (TD) children (age 7-12 years, IQ>70) were measured with 64- channel EEG in an eyes-closed resting-state condition. All EEG recordings were visually scored for (epileptiform) abnormalities. We quantified the EEG oscillations using spectral and temporal methods in the standard frequency bands and averaged values across channels within three large regions of interest. We also characterized long-range temporal correlations of ongoing oscillations as a proxy of E/I balance using detrended fluctuation analysis (DFA). Integration of biomarkers was done in the Neurophysiological Biomarker Toolbox (http://www.nbtwiki.net/), employing data-mining algorithms to combine information from multiple biomarkers into a single index.


We found a high number of EEG abnormalities within the ASD groups. ASD patients had high delta relative power and strong autocorrelations in widespread brain regions: both ASD and ASD-EPI subjects showed elevated long-range temporal correlations compared with TD, suggesting that the E/I balance is indeed disturbed in ASD (Poil et al., 2012). The machine-learning techniques resulted in integrated EEG-biomarker indices that discriminated the two ASD groups from TD with high accuracy, and predicted the EEG abnormality scores. The EEG index that optimally discriminated ASD and TD children showed high correlation with total IQ, Sensory Profile scores and Social Responsiveness Scale. The index discriminating ASD-EPI from ASD correlated highly with Autism Behaviour Checklist-Irritability subscale.


Our findings suggest that these EEG markers can be used to stratify ASD patients, which might contribute to the development of a decision support system for diagnostic and prognostic use in the clinical assessment of children with various manifestations of ASD.