Data Preprocessing and Quality Control

Data Extraction

The data shared in this project are available as raw data, but also preprocessed. The data from each paradigm is saved as a separate file. See here for instructions on accessing the preprocessed data, as well as the MATLAB code for the preprocessing.

Electrode Quality Check

In preprocessed data, bad electrodes were identified and replaced. Identification of bad electrodes was based on probability, kurtosis, and frequency spectrum distribution of all electrodes. A channel was defined as a bad electrode when recorded data from that electrode exceeded more than 3 standard deviations from all other electrodes. This was realized with the eeglab MATLAB function: “pop_rejchan.m”. Subsequently, bad electrodes were interpolated using a spherical spline interpolation (Perrin et al., 1989, 1990) ‘eeg_interp.m’. Moreover, after automatic scanning, noisy channels were selected by visual inspection and interpolated or replaced entirely by zeros (for the calculation of the ISC measures to eliminate the channel’s contribution in subsequent calculation of covariance matrices).

Artifact Signal Correction

One hundred nine EEG channels were used for scalp recordings, while 6 EOG channels were used for artifact removal. The rest of the channels, mainly neck and face channels, were discarded before data analysis. Data were then high-pass filtered at 0.1 Hz and notch filtered at 59-61 Hz. Eye artifacts were removed by linearly regressing the EOG channels from the scalp EEG channels. Next, a robust Principal Components Analysis (PCA) algorithm, the inexact Augmented Lagrange Multipliers Method (ALM, Lin, Z., Chen, M., & Ma, Y., 2010) removed sparse noise from the data. Briefly, the ALM recovers a low-rank matrix, A, efficiently and accurately from a corrupted data matrix D = A + E, where some entries of the additive errors E may be arbitrarily large. Finally, the entire data set for each subject was visually inspected in order to discard whole block and/or paradigm recordings that remained noisy after the automatic and manual noise removal methods. All signal processing was performed offline using MATLAB software (MathWorks, Natick, MA, USA).