Fig 1.
An example of muscular artifact suppression using the ICA.
(A) The original EEG signals x[n] corrupted by EMG artifacts. (B) The separated source components , which were classified as brain signals (denoted as EEG) and muscular artifacts (denoted as EMG). (C) The reconstructed EEG signals
with suppressed muscular artifacts.
Fig 2.
An example of a failed attempt to separate muscular artifacts from 111 channel EEG record.
(A) EEG signals x[n] corrupted by EMG artifacts (for spatial reasons only 25 out of 111 EEG channels are shown). (B) The source components provided by ICA (again for spatial reasons only 25 out of 111 components are shown; the remaining components look very similar to those that are presented—they are composed of occasional bumps, and no clear separation of EEG and muscular artifacts can be recognized).
Fig 3.
A histogram of power ratios α for source components classified as EMG (white) and source components classified as brain signals (black).
Fig 4.
An example of EEG with the simulated EMG contamination.
(A) 25 EEG channels with simulated EMG contamination. (B) Topographic maps of average power spectral density in various frequency bands.
Fig 5.
Examples of PSDs of EEG with simulated EMG contamination at several scalp locations.
Fig 6.
Mean correlation coefficients between the original clean EEG signals and EEG signals with suppressed muscular artifacts as a function of subspace dimension L for 111, 10/10 and 10/20 electrode systems.
Table 1.
The correlation coefficients from the processing of EEG signals without muscular artifacts (ξ = 0).
Table 2.
The correlation coefficients from the processing of signals with equally strong muscular artifacts and EEG (ξ = 1).
Table 3.
The correlation coefficients from the processing of signals with muscular artifacts four times stronger than EEG (ξ = 4).
Fig 7.
An example of EEG signals processed by the newly proposed method.
The original signals are shown in Fig 2A. The EEG contains 111 channels, but for spatial reasons we show only 25 channels that correspond to the ones shown in Fig 2A.
Fig 8.
Topographic maps of the average power spectral density in various frequency bands before and after processing by the proposed algorithm.
Fig 9.
Power spectral densities at several scalp locations before and after processing by the proposed algorithm.
Fig 10.
Average PSDs showing changes in alpha power band for eyes open and closed.
(A) Average PSDs for resting EEG measured with closed eyes. (B) Average PSDs for resting EEG measured with open eyes. The black lines are PSDs computed from the unprocessed data, while the red ones are PSDs computed from the data processed by the proposed algorithm.