Preprocessing and artifact rejection

Scalp electroencephalographic (EEG) channel space data preprocessing was conducted using EEGLAB (Delorme and Makeig, 2004) through a procedure outlined in Figure ​Figure1A.1A. EEG data from each of the 10 participants were band-pass filtered, retaining frequencies from 1 to 55 Hz, and re-referenced to the channel average since the data was recorded reference-free. The data were epoched, extracting the time period of interest, which encompassed both the planning phase and the execution phase of every trial. No distinction was made between reach directions since the emphasis for this analysis was placed on the general process of reach/saccade planning and execution rather than directional discrimination and, as such, both upward and downward angled reach trials were pooled and analyzed together. The control trials (Lift) and test trials (Reach/Saccade) were analyzed separately, using a cortical network defined over the same ROIs. Differences between the control and test conditions were examined by considering differences in network dynamics between the control and test conditions during the planning and execution phases. Noise and artifact laden trials and channels were identified and removed in a semi-automated manner. The kurtosis of each channel or trial was used as the identification metric, with channels or trials exhibiting a kurtosis Z-score > +5 or < −5 being flagged and subsequently visually inspected before removal.

The epoched channel space data were decomposed into sets of maximally independent components, where independence was achieved by minimizing mutual information between components using Infomax ICA (Bell and Sejnowski, 1995) (process outlined in Figure ​Figure1B).1B). These components represented a combination of putative effective cortical sources, muscular artifacts, ocular artifacts, and electrical activity from the heart, among other noise sources. Through a semi-automated procedure, cortical sources were defined by employing independent components that both visually, by identifying power spectra and topographical maps with dipolar cortical source character, and computationally, through dipole-fitted residual variance, corresponded to dipoles (each fitted to a dipole with low residual variance). This procedure is particularly well suited for EEG artifact removal since the electrophysiological activity of interest present in the cortex manifests detectably in the form of an electrical dipole. The linear mixing of the electric fields generated by these dipoles, and corresponding unmixing through ICA allows for very precise retention of cortical data and rejection of non-cortical artifacts. Using the weight matrix computed during ICA, the retained cortical independent components were re-projected to channel space, creating cleaned channel space electrical signals that were further downsampled to 128 Hz for use in subsequent processing.