The Impact of Child Characteristics on EEG Data Quality in Infants at Risk for and Children with ASD

Background:  

Electroencephalography (EEG) is a promising functional imaging method to identify biomarkers of outcome in children at risk for and with ASD. EEG data are susceptible to artifact and can be rendered unusable due to slight movements, such as head turning, blinking, or even smiling. During initial stages of data processing, artifacts commonly result in trial loss. Currently, no standardized measures exist to quantify participant behavior and its effects on data quality. Additionally, it is undetermined if variability in social communication impairment and cognitive function impacts data quality. 

Objectives:  

We asked whether cognitive ability and behavior of children with or at risk for ASD related to percentage of useable trials garnered from session. We developed a rating of child’s behavior during session through a 5-point likert scale of perceived mood. Primary use of the scale was to quantify in-session perceived mood and overall compliance. We investigated two samples of children from the Autism Centers for Excellence (ACE) study. The two groups included: (1)“ACE Project 2:” 12-24 month-old infants at risk for ASD, defined by elevated scores on the Autism Diagnostic Observation Schedule-Toddler Version (ADOS-T) and (2) “ACE Project 3:” children aged 6-11 years-old with ASD who are minimally verbal. We focused on these cohorts due to marked cognitive impairment and varied symptom expression, resulting in challenges during EEG recording. 

Methods:  

Forty-seven children from ACE Project 2 (mean age=18.6 months) were presented with familiar and novel photos of faces while EEG was recorded. Sixteen children from ACE Project 3 (mean age=88.2 months) listened to an auditory statistical learning task while EEG was recorded. EEG recording was performed using a high-density system (128-channel, EGI Inc.). Developmental quotients (DQ) were acquired from the Mullen Scales of Early Learning and social communication impairment was calculated with the total score from the ADOS-T and ADOS-2. Behavior rating was performed during EEG tasks. Useable data were defined as the percentage of trials available after automated artifact detection; channels were rejected if amplitude difference was greater than 150 mV.

Results:  

ACE Project 2 and ACE Project 3 participants did not differ in amount of useable data provided (t=1.36,p=.262). Within ACE Project 2, age (r=.128,p=.509), ADOS-T score (r=-.12,p=.536), and DQ (r=-.083,p=.668) did not relate to amount of usable data collected from participants. Similarly, within ACE Project 3, age (r=.200,p=.950), ADOS-2 score(r=.074,p=.747), and DQ (r=-.465,p=.128) did not significantly relate to useable data. Behavioral rating significantly correlated with amount of useable data in both groups [ACE Project 2 (r=-.404,p=.030) and ACE Project 3 (r=-.734,p=.01)].

Conclusions:  

Clinical characteristics, including cognitive ability and social communication impairment, are not useful constructs to measure data quality in EEG. However, perceived mood and behavior of the child during session does predict useable data. As EEG and other experimental measures serve as functional biomarkers in ASD, measurement of a child’s behavior and mood during testing sessions will be critical to improve signal quality. Practices to standardize behavior and improve mood during recording should be implemented. Future directions involve measuring impact of behavioral regulation techniques on data quality.