Artifact Rejection
We implemented a fully automated artifact rejection framework (for
details, see (Murty et al. , 2020, 2021; Murty & Ray, 2022), in
which outliers were detected as repeats with deviation from the mean
signal in either time or frequency domains by more than 6 standard
deviations, and subsequently electrodes with too many (30%) outliers
were discarded. Subsequently, stimulus repeats that were deemed bad in
more than 10% of the other electrodes (see below) or in the visual
electrodes were discarded. This gave rise to a set of good electrodes
and common bad repeats for each subject. Overall, this led us to reject
15.33±0.36% (mean±SEM) stimulus repeats. We also rejected electrodes
with high impedance (>25kΩ). Finally, as in our previous
studies, we computed slopes for the baseline power spectrum between
56-84Hz range for each unipolar electrode and rejected electrodes whose
slopes were less than 0. Based on these criteria, 15.30±0.72%
(mean±SEM; min: 0, median: 12.5%, max: 60.9%) electrodes were labelled
bad. Some subjects who had predominantly bad electrodes
(>40 electrodes) were rejected later in the pipeline.
Eye position was monitored using EyeLink 1000 (SR Research Ltd), sampled
at 500Hz. Any eye-blinks or change in eye position outside a 5º fixation
window during -0.5s to 0.75s from stimulus onset were noted as fixation
breaks, which were removed offline. This led to a rejection of
16.54±0.059% (mean±SEM) stimulus repeats. Overall, we rejected
31.88±0.81% (mean±SEM) of the repeats (min: 8.4%, median: 30.1%, max:
73.2%).