Techniques of EEG Recording and Preprocessing - Quantitative EEG Analysis Methods and Clinical Applications - page 44

topographies of the 22 independent components. The eye movement artifact
(between seconds 2 and 3) was isolated by ICA to components IC1, IC5, and IC21.
The scalp maps indicate that these components account for the spread of EOG activ-
ity to frontal sites. Components IC13, IC14, and IC15 evidently represent muscle
noise from temporal muscles. After eliminating these artifactual components, by
zeroing out the corresponding rows of the activation matrix u and projecting the
remaining components to the scalp electrodes, the “corrected” EEG data [Figure
2.8(c)] are free of both EOG and muscle artifacts. Removing EOG activity from
frontal channels reveals alpha activity near 8 Hz that occurred during the eye move-
ment but was obscured by the eye movement artifact in the original EEG traces.
Close inspection of the EEG records [Figure 2.8(a)] confirms its presence in the raw
data. The artifact-corrected data also reveal underlying EEG activity at temporal
sites T3 and T4 [Figure 2.8(c)] that was well masked by muscle activity in the raw
data [refer to Figure 2.8(a)].
The second example (Figure 2.9) demonstrates that ICA can also be used to
remove stimulus-induced eye artifacts from unaveraged event-related EEG data
through analysis of a sample data set collected during a selective attention task. The

+

0
−

(a) (b)

HEOG VEOG HEOG VEOG

Fz Fz

FC1 FC2 FC1 FC2

C3 CZ C4 C3 CZ C4

CP1 CP2 CP1 CP2

Pz Pz

+ 15

μ V

Oz Oz − 10

− 100

900

Time (ms)

(c) (d)

Figure 2.9 Elimination of eye movement artifacts from ERP data. (a) The scalp topography of an ICA
component accounting for blink artifact. This component was separated by ICA from 555 target
response trials recorded from a normal subject in a visual selective attention experiment. ( Note:
Because this scalp map interpolation was based on very few frontal electrodes, it is not a representa-
tive depiction of an eye blink component map.) (b) The scalp map of a second component account-
ing for lateral eye movements. (c) Averages of ( N = 477) relatively uncontaminated and ( N = 78)
contaminated single-trial target response epochs from a normal control subject. (d) Averages of
ICA-corrected ERPs for the same two trial subgroups overplotted on the average of uncorrected
uncontaminated trials.