## Similar protocols

## Protocol publication

[…] For ICA, continuous EEG data were first imported into **EEGLAB** 12.0 (Delorme and Makeig, ), which is an open-source software toolbox (Swartz Center for Computational Neuroscience, La Jolla, CA; http://sccn.ucsd.edu/eeglab/) that was run in Matlab version 8.0 (The Mathworks, Inc., Natick, MA). A 38-channel subset of electrodes was selected for further analysis. The choice of 38 sites was based on having an adequate sampling of spatial scalp sites while also limiting the number of channels and computation time. The 38 electrode sites were equally distributed within frontal to posterior (FP to O) and medial to lateral (z to 7/8) sites based on the 10–20 system. The EEG was high-pass filtered at 1 Hz, re-sampled at 250 Hz, and re-referenced to the average of all 38 scalp channels. The resulting continuous data were segmented into epochs from 1200 ms before to 1200 ms after stimulus onset. A longer epoch was used for the ICA analysis because ICA solution strength (i.e., amount of mutual information reduction between components) increases when applied to a larger number of time points and because accurate time-frequency analysis at lower frequencies (<5 Hz) requires a longer time window. The data were visually inspected for the presence of outlying data and non-stereotyped artifacts. Channels containing non-stereotyped artifacts throughout the recording (e.g., line noise) and epochs containing other non-stereotyped artifacts (muscular potentials, scalp-electrode connectivity or movement) were removed from the data.After the data rejection procedure there was a mean of 35 ± 1 channels and 332 ± 11 epochs per participant. We then performed extended infomax ICA on the data (Bell and Sejnowski, ). Independent component analysis finds an unmixing square matrix with rows and columns equal to the number of input channels which, when matrix-multiplied with the raw data, provides maximally temporally independent activations. Each independent component (IC) activation has a fixed topographic projection map to scalp channels which is given by the inverse of the unmixing matrix. Independent components for each participant were accepted for further analysis based on scalp map topography (smooth regions of positive and negative polarity that are well-distributed across channels), mean log spectrum (1/f-like curve with typical EEG spectral peaks, e.g., θ, α, β frequency bands), and consistent trial-to-trial activations as evidenced by time-locked peaks in epoch averages (ERPs) and corresponding peaks in trial activations. Independent components having characteristics indicative of stereotyped artifacts (eyeblinks, eye movements, electrocardiogram, and muscular potentials) were removed. A total of 512 ICs (mean = 15 ± 1 per participant) were selected for further analysis.The neural source of the selected ICs was modeled using DIPFIT2 using functions from the **FIELDTRIP** toolbox [(Oostenveld et al., ); Donders Institute for Brain, Cognition and Behavior; http://fieldtrip.fcdonders.nl/]. DIPFIT2 uses scalp topographic maps as an input, and calculates the location and orientation of a single equivalent current dipole using a three-shell boundary element model. A standard boundary element head model was used for all participants and was composed of three 3-D surfaces (skin, skull, cortex) extracted from the Montreal Neurological Institute (MNI) canonical template brain. Scalp channel locations were co-registered with locations in the model space by aligning them with their standard locations in the 10–20 system relative to the MNI template.Time-frequency analysis of the IC activations, known as the event-related spectral perturbation (ERSP), was also calculated. The ERSP visualizes mean event-related changes in spectral power over time in a broad range of frequency bins. In doing so the ERSP generalizes classic event-related desynchronization and synchronization measures (Pfurtscheller and Aranibar, ). Time-frequency analysis of single-trial IC activations was performed by convolving the data with a Morlet wavelet that used 3 cycles at the lowest frequency (2.5 Hz) and a linearly increasing number of cycles up to 30 at the highest frequency (50 Hz). The result was scaled to decibels (dB), and the values in the post-stimulus period were normalized for each frequency by subtracting the mean value in the baseline period. The ERSP was imaged by plotting the normalized power values as a color within a “heat map” in a 2-D time-frequency plot.A clustering procedure was used to determine which ICs from different participants represented similar functionally distinct EEG processes. We first pre-defined and computed four measures for each IC: scalp map, dipole location, ERP, and ERSP. For each measure a data space was constructed in which measures could be compared across all ICs. Principal component analysis was applied to separate the IC data along a pre-defined number of orthogonal dimensions based on both spatial (scalp map and dipole location) and non-spatial (ERP and ERSP) features. The resulting principal component templates were concatenated and principal component analysis was then applied to reduce the total number of dimensions by half. This resulted in a set of data in which each component possessed a value in a 15-dimensional data space. We then applied a k-means clustering algorithm to this data space which separates components into k clusters and observed the results for k = 8 to k = 15. During the clustering process, we also removed components whose centroids were >3 SDs from the centroid of any cluster metric space. Based on the consistency of clustering solutions for increasing values of k, we identified six clusters of interest that had both a high proportion of participant contribution and whose activities contained significant effects. If a subject contributed more than one component to a given cluster two of the authors (JM and MS) independently chose one component per participant within each of the six clusters of interest based on assessment of the measures used for clustering. The two independent sets of IC placements agreed on 459 out of the 512 total ICs (89.6%). For the remaining 53 disagreements, the two authors compared the relevant ICs and came to an agreement for the final clustering. […]

## Pipeline specifications

Software tools | EEGLAB, FieldTrip |
---|---|

Application | Clinical electrophysiology |

Organisms | Homo sapiens |

Diseases | Hypoplastic Left Heart Syndrome |