Comparison and Recommendations for Processing EEG Data in Children with Autism and Typical Developing Controls

Background: A particularly useful and increasingly utilized tool for studying children with ASD is quantitative electroencephalography, qEEG, which allows for analysis of the power at different established frequency bands. qEEG, as a biomarker, has the potential to define subgroups of children with ASD, which can then inform targeted treatment. However, different procedures for processing qEEG data can have a major impact on final power measurements, which leads to difficulties in replicating findings. In order to maximize the utility and reliability of qEEG, a comparison of different processing methods is needed.

Objectives: We have investigated the stability and reliability of qEEG in typically developing (TD) children, the way in which different decisions in the data processing stream affect the data, and whether these parameters differ between children with ASD and TD controls. This project addresses many methodological questions and compares decisions at various processing steps. Some of the main questions we seek to answer include: 1) What is the minimum amount of clean EEG data required to produce a stable measurement of the power in well-known frequency bands (theta, alpha, beta, and gamma); 2) How stable are the measurements of these frequencies across the course of the single recording session (~5 minutes), across multiple time points (~3 months), and at different ages (2.5 – 6 years old); 3) How does the selection of the EEG reference and electrode groupings affect the measurements of these frequency bands; and (4) How do the answers to these questions differ between children with ASD and age-matched TD controls.

Methods: EEG was recorded while children were at rest with their eyes open, watching a video of bubbles. A minimum of 2 minutes of data were obtained from both children with ASD and TD controls. EEG data were bandpass filtered from 1 to 50 Hz, divided into one second segments, and examined for artifact contaminated data. Various crucial decision points in a typical qEEG processing stream were altered and compared for their impact on standard statistical measurements of frequency power. After measureable changes in the EEG data processing stream were made, the data were transformed into the frequency domain using a Fast Fourier Transform (FFT) in order to obtain a measurement of the Power Spectral Density (PSD) at standard frequency bands.

Results: Given the large number of experimental manipulations and comparisons, all results are too numerous to summarize. However, a few brief, but important findings include the following. 1) The number of segments used to calculate frequency power differs across frequency bands. 2) Different artifacts can have very different effects on each frequency band. 3) Selection of electrode groupings must be performed with great care.

Conclusions: The decisions that a researcher makes on how to process and clean their qEEG data greatly affects the results. If qEEG is to become a useful biomarker across labs, then standardized procedures should be followed. Based on our many comparisons, we provide recommendations for others who use high-density electrode systems to measure neural activity in children with ASD and TD children.