Low Density EEG Files from Bucharest Early Intervention Project for HAPPILEE Software

The various steps of the HAPPILEE automated pipeline (for low-density electroencephalography (EEG)) were optimized using a subset of developmental EEG files from the Bucharest Early Intervention Project (BEIP) (for full study design, see Zeanah et al., 2003). The optimization dataset includes resting-state EEG from three groups of children living in Romania starting in 2001. The first group, referred to as the Care as Usual Group (CAUG), is composed of children living across six institutionalized care facilities throughout Bucharest, Romania. The second group is the Foster Care Group (FCG), which is composed of children who were removed from these institutions through random assignment and placed in a foster care intervention. The final group is the control group (CG), made up of a community sample of children living with their biological families who have never been placed in institutionalized care or foster care. We selected a subset of thirty EEG files across the three groups from the greater dataset (CAUG, n=8; FCG, n=8, CG, n=14). The average age at the start of the baseline assessment across the three groups was 17.40 months, with a range of 6.28 - 29.98 months (averages per group: CAUG=18.30; FCG=16.21; CG=17.58). The resting-state EEG for all children was recorded with the James Long system from twelve scalp sites (F3, F4, Fz, C3, C4, P3, P4, Pz, O1, O2, T7 and T8) using a lycra Electro-Cap (Electro-Cap International Inc., Eaton, OH) with sewn-in tin electrodes.

The thirty full-length files from the BEIP EEG dataset are titled A.set, B.set, etc. There are three 30-second EEG segments per subject (45 total) with the following conditions: The _artifact_laden files have data with heavy amounts of artifact, the _clean.set files have data determined to be generally clean and without considerable artifact, and the _artifact_added.set files contain the clean 30-second data segments with artifacts taken from the same subject added to the clean data. To create the _artifact_added.set files, we isolated artifact timeseries by running ICA on the artifact laden 30-second files and selecting approximately 2 components that were determined to be artifact with minimal neural data via visual inspection and automated classification through both the ICLabel and Multiple Artifact Rejection Algorithm options. We subsequently added those artifact component timeseries on top of the clean 30-second data segment from the same individual. These three types of files were used to evaluate the HAPPILEE pipeline performance for artifact removal.

The simulated_data.set EEG data was created with code from Bridwell et al. (2018). In short, the simulated EEG consisted of four signals. The four signals had distinct spatial patterns and frequency ranges (1.00-3.91Hz, 3.91-7.81Hz, 7.81-15.62Hz, and 15.62-31.25Hz). For a more thorough description on how the simulated signals were created, see Bridwell et al., 2018. The simulated data is high-density EEG, but a subset of channels was selected for analyses to mimic low-density data constraints. The simulated_signal_with_artifact.set contains the simulated EEG data with blink and muscle artifact added. To get the blink artifact, we used a clear blink independent component (IC) from an adult participant (see Leach et al., 2020 for the specific study details). This IC was selected based on both an automated artifactual IC detection algorithm and visual inspection by two researchers with over five years of EEG and at least two years of ICA experience. For the muscle artifact, we pulled eight muscle ICs from the BEIP dataset used above in the artifact addition approach. After adding the artifact to the simulated EEG data, we did a 1Hz highpass and 35 Hz lowpass filter. Following this, we epoched the data into two-second epochs (50% overlap) to prepare the data for wavelet thresholding and/or artifact rejection. For artifact rejection, we used a -100 to 100 μV voltage threshold to identify bad epochs. We also required both frontal electrodes to exceed this threshold in order to classify an epoch as containing a blink.

The Bucharest Early Intervention Project was carried out in accordance with the recommendations of the Institutional Review Boards at Boston Children’s Hospital, University of Maryland, and Tulane University. Moreover, informed consent was signed by the Commissioner for Child Protection for each child participant living in his sector of Bucharest, as dictated by Romanian law. Further assent for each procedure was obtained from each caregiver who accompanied the child to the visit.

All files here have been deidentified.