Automatic determination of EMG-contaminated components and validation of independent component analysis using EEG during pharmacologic paralysis

• We present a conceptually simple, spectral-based, heuristic to automatically identify independent components that are predominantly EMG.
• Spectra with slopes above a certain threshold are used to define EMG-contaminated components, which are then removed.
• We validate its effectiveness using EMG-free data recorded under pharmacological neuromuscular paralysis.

ObjectiveValidate independent component analysis (ICA) for removal of EMG contamination from EEG, and demonstrate a heuristic, based on the gradient of EEG spectra (slope of graph of log EEG power vs log frequency, 7–70 Hz) from paralysed awake humans, to automatically identify and remove components that are predominantly EMG.MethodsWe studied the gradient of EMG-free EEG spectra to quantitatively inform the choice of threshold. Then, pre-existing EEG from 3 disparate experimental groups was examined before and after applying the heuristic to validate that the heuristic preserved neurogenic activity (Berger effect, auditory odd ball, visual and auditory steady state responses).Results(1) ICA-based EMG removal diminished EMG contamination up to approximately 50 Hz, (2) residual EMG contamination using automatic selection was similar to manual selection, and (3) task-induced cortical activity remained, was enhanced, or was revealed using the ICA-based methodology.ConclusionThis study further validates ICA as a powerful technique for separating and removing myogenic signals from EEG. Automatic processing based on spectral gradients to exclude EMG-containing components is a conceptually simple and valid technique.SignificanceThis study strengthens ICA as a technique to remove EMG contamination from EEG whilst preserving neurogenic activity to 50 Hz.