Analysis of EEG Signals

The filtered EEG signals were first decomposed into independent brain sources by

independent component analysis (ICA) (Bell & Sejnowski, 1995; Makeig et al., 1997)

using EEGLAB (Delorme & Makeig, 2004). The ICA algorithm can separate N

source components from N channels of EEG signals. The summation of the EEG

signals at the sensors is assumed to be linear and instantaneous, i.e. the propagation

delays are negligible. We also assume that the time courses of muscle activity, eye,

and, cardiac signals are not time locked to the EEG activities reflecting synaptic

activity of cortical neurons. Therefore, the time courses of the sources are assumed to

be statistically independent. The multi-channel EEG recordings are considered as

mixtures of underlying brain sources and artifacts. The source signals contribute to

the scalp EEG signals through a fixed spatial filter. Such a spatial filter can be

reflected by the rows of inverse of unmixing matrix, W in u = Wx, where u is the

source matrix and x is the scalp-recorded EEG. The spatial filters can be plotted as the

scalp topography of independent component. The scalp topography of each

independent component (IC) can be further analyzed using DIPFIT2 routines

(Oostendorp & Oostenveld, 2002), a plug-in in EEGLAB, to find the 3D location of

25

an equivalent dipole or dipoles based on a four-shell spherical head model. Then,

components with similar scalp topographies, dipole locations and power spectra from

many subjects were further grouped into component clusters to examine the

consistency of brain areas involved in the task. Ten component clusters recruited more

than 10 components from multiple subjects with similar topographic maps. Among

these robust component clusters, we further correlated the component power spectra

with subjects’ continuous motion-sickness rating. Average correlation coefficient was

computed for each of the robust component clusters. Then, the five most motion

sickness level-related clusters were selected for further analysis.

Relationship between spectra and road-condition or motion-sickness

As mentioned above, brain signals can be sensitive to any environmental change.

Hence the EEG signals acquired under various conditions (such as on a straight or

curved road) should not be confounded by motion-related activities. For example,

when the experiment entered the winding-road riding section, the car began to sway

left and right with the VR scene of the curved road, providing both visual and body

sensation stimuli to the subjects. This baseline difference among EEG power spectra

associated with the different road conditions must be considered when the MS-related

EEG power changes are evaluated.

26

A

B

Fig. 2-6. Road-condition and motion-sickness effects can be extracted from the

signals by comparing EEG signals in the red and blue blocks.

Therefore, the EEG power spectral changes in three periods were initially examined:

(1) baseline - the first 200 seconds of the baseline straight road section, (2) low MS

level - the first 200 seconds of the curved road section, and (3) high MS level - the

first 200 seconds after the highest sickness rating (Fig. 2-6 B). The power spectra in

these three time periods (baseline, low-sickness and high-sickness) were then

averaged among subjects in each selected IC cluster.

A statistical analysis was also conducted to assess the significance of the spectral

differences of the independent components under different motion-sickness levels and

various road conditions. Since the true sample distribution of the component spectra

27

was unknown and the sample size (n=19 as 5 of 23 subjects were excluded due to low

quality of EEG data) was small, a nonparametric statistical analysis, a paired-sample

Wilcoxon signed-rank test, was employed to access the statistically significant

spectral differences between different conditions. The level of significance was set to

p <0.01.

Time-frequency Analysis

Time-frequency analysis was utilized to test the dynamics of the ICA power

spectra throughout the experiment. The time series of the ICA power spectra were

then correlated with the continuous sickness-level to determine the MS-related

spectral changes. The frequency responses of ICA activations were calculated using a

500-point moving window with 250 overlapping points. The 500-point epochs were

further subdivided into several 125-point sub-windows with 25-point overlaps. The

125-point sub-windows were zero-padded to 512 points to calculate the power spectra

using a 512-point fast Fourier transform (FFT), yielding an estimate of the

power-spectrum density with a frequency resolution of 0.5 Hz. The power spectra of

these sub-windows were then averaged to produce a power spectrum for each 2 s

epoch. The power spectrum density (PSD) was then converted into dB power. The

temporal resolution of the resultant spectral time series was 1 s since the window step

28

was 250 points and the EEG was sampled at 250 Hz. The analysis procedure is shown

in Fig. 2-7.

It is possible that both road condition (i.e., the big sway of the VR driving model)

and motion-sickness level can simultaneously alter EEG power spectrum. Here, a

two-stage baseline removal was used to dissociate the baseline EEG power spectrum

associated with different road conditions. First, we computed the baseline EEG power

of the baseline section by averaging the first 3-minute EEG power spectrum of the

baseline section and subtracted this baseline EEG power from the entire EEG signal.

Second, we computed the baseline EEG power from the first 3 minutes of EEG

signals acquired during the motion-sickness section and subtracted this baseline EEG

power only from the EEG signals of the motion-sickness section. We assumed that the

EEG power changes caused by the moving platform would be consistent through the

entire motion-sickness section and, thus, we could extract mainly the motion-sickness

related EEG power changes by removing the baseline EEG power derived from the

beginning of this section. However, there might be interaction between the level of

movement of the platform and motion-sickness level. This remains a question for

further exploration. In this study, we tried to program the movement of platform as

consistent as possible through the motion-sickness section and kept this parameter

well controlled.

29

Fig. 2-7. Time-frequency analysis procedure used to obtain dynamic EEG frequency responses during the experiments.

Fig. 2-8. Baseline removal of the time-frequency results.

30

MS-sorted EEG spectra

To examine the relationship between the severity of motion-sickness and

concurrent changes in the EEG spectrum, the EEG power spectra were first computed

in the winding-road section, and the spectra were sorted by subjective MS level for

each of the component. The sorted EEG spectrogram at each MS level was then

averaged across the components within each IC cluster. Linear regression was used to

determine if the spectral changes vary as a function of MS level.

Time Relationship among MS-related EEG Processes

Multiple components exhibited MS-related spectral changes. The time relationship

between these EEG processes is of interest. A cross-correlation analysis was

performed on the power spectra of selected ICs for each individual. To be more

specific, the spectrogram at each frequency were temporally shifted from -200 to 200

s with a step size of 1 s and correlated with the time series of the subject's MS ratings.

Finally, the results were averaged across components in each of the five IC clusters.