Event Related Spectral Perturbation (ERSP) Analysis

Chapter 3 Data Analyses

3.4 Event Related Spectral Perturbation (ERSP) Analysis

The ERSP, a kind of time-frequency analysis, which was first proposed by Makeig (Makeig, 1993), can reveal those time-locked but not necessary phase-lock event related activities. ERSP analysis transforms time-course signal into spectral-temporal domain. Therefore, activities of each frequency band associated

with navigation can be further analyzed. Fig. 3-7 illustrates an example of ERSP analysis of left motor component.

Figure 3-6: The picture shows the illustration of procedures in ERSP analysis. Short term FFT is applied in each window with 256 sample points, and overlaps by 244 sample points. In the final step, non-significant parts of ERSP image are set to zero by the means of bootstrap.

The processing flow was shown in Fig. 3-6. ICA activation was first divided into 200 512-point windows. Each window was multiplied with Hanning gain and then extended to 1024 points by zero-padding to calculate its power spectrum fast Fourier transform (FFT). Log power spectra were computed and then were normalized by

subtracting the baseline (straight tunnel segment) log mean power spectral. This procedure was applied to all the epochs. Time series of IC in each trial were transformed into time-frequency matrix (200 x 1024) with a frequency resolution about 0.05 Hz the results were then averaged to yield ERSP image. Fig. 3-7 gave an example of ERSP image which is motor component. Significance of deviations from power spectral baseline was assessed by bootstrapping, a nonparametric permutation-based statistical method. Non-significant points were masked as zero;

only significant (p<0.05) perturbations were remained. Through ERSP, we investigated the time-frequency information of brain activity in different area during the tunnel task.

Figure 3-7: Pictures show the illustration of ERSP images of the left somatomotor component. (A): ERSP images of the left somatomotor component which do not mask by significance level. Note: the powers at around 10- and 20- Hz frequency bands shows apparently suppressed during the selecting homing angle period in comparison to the baseline period. Pink dashed lines: the averaged onset of forthright. Blue dashed lines: the mean onset of meander. Red dashed lines: the averaged response time of the homing direction selections. Black dashed lines: the mean response time of the homing angle selections. Color bars showed the power magnitude of ERSPs.

(B) The same ERSP images shows in (A) but masks by significance level. The significance level was set at p<0.05.

3.6 Component Clustering

To compare electrophysiological results across subjects, the usual practice of most researchers has been to identify scalp channels. Component clustering would be used to assess the consistency of ICA decompositions across subjects and sessions, and to evaluate the separate contributions of identified clusters of these data components to the recorded EEG dynamics.

ICA components from multiple subjects and sessions were clustered semi-automatically based on their gradient of scalp projections, EEG characteristic and equivalent dipole location. These criteria for each IC were measured and compressed into 10-dimensional feature direction by principal component analysis (PCA). Semi-auto K-means algorithm was used to group similar ICs together and the clustering result was finally adjusted according to the time-frequency property and equivalent dipole location of each IC. The grand mean scalp map and spectra property were then computed for each cluster to investigate the common characteristics brain activity in the task. The process of clustering was shown as Fig. 3-8. Further analysis will base on the clustering result.

Components from 1 subject Components from 1 subject

Figure 3-8: The picture shows the flowchart of component clustering analysis using the K-means algorithm. Brain process associated independent components (ICs) across subjects and sessions are first selected by observations according to theirs scalp map distribution and spectral property. Then, the selected ICs are semi-automatically clustered by Kmeans algorithm into several clusters which had similar EEG

characteristics. These clusters represent the activities from different brain areas, such as parietal, occipital, left somatomotor and right somatomotor area.

Chapter 4 Result

A total of 20 subjects completely finished this experiment; 11 of them were categorized into allocentric subjects, and 7 participants were categorized into egocentric subjects. The rest three subjects’ were not able to be classified into any category because their homing direction selections varied trials by trials. The influences of the choice of reference frame during the navigation on subjects’

behavioral performance were presented in the first section. Then, the characteristic neural activities among the different brain regions in these two groups were presented in the second section and the last section presented differences on the brain oscillations in the parietal and the occipital regions between these two types of subjects in terms of the baseline and phasic power spectra as well as the ERSPs.

4.1 Behavioral Performance

The mean response time for determining the homing direction of 6 groups were ranged from 550 to 650 msec (Fig. 4-1). No apparent differences reflected on distributions of the response time of the homing directional selection between the allocentric and egocentric subjects [F(1,2256) = 0.2324; p = 0.79265 > 0.05].

Similarly, the turning degree did not have any influences on the speed of the homing directional selections [F(2, 2256) = 0.8768; p = 0.34918 > 0.05]. No interactions were found between the factors of strategy and turn eccentricity [F(2, 2256) = 0.56369; p = 0.56918 > 0.05].

Figure 4-1: Effects of the navigation strategy and the turning degree on the grand mean of the response time for determining homing directions. Note no differences in the distribution of the response time are observed among 6 different groups.

Effects of the navigation strategy and the turning degree on the angle fit are showed in Fig. 4-2. The average absolute errors of the homing direction were lower than 15° for the tuner and non-tuner subjects in three different turn angle. To further test the significance of effects on turning degree and the navigation strategy, we use 2 x 3 repeated-measures ANOVA. Results showed both the navigation strategy and turn angles have significant influences on the absolute errors of the angle fit [navigation strategy: F(1, 2256)=98.993; p < 0.001; turning degree: F(2, 2256)=457.865; p <

0.001), but no interactions were found between the two factors [F(2, 2256)=0.876; p

>0.05]. For the allocentric reference frame users, the correctness for pointing the homing direction were decreased with the increases of the turn angle but such sequential drop on the response accuracy along with the increases of the turn angles did not reveal in the egocentric reference frame users. Specifically, the response

accuracy were as 60° > 30°> 90°. Comparing with the allocentric reference frame users, the response accuracy were significant higher for the egocentric reference frame users when turn angles were larger (60°: 7.52 ± 0.29 < 8.22 ± 0.22, p<0.005;

90°: 11.06 ± 0.37 < 14.32 ± 0.41, p<0.000) but their response accuracy were significant lower when the turn angles were small (30°: 5.39 ± 0.19 < 9.45 ± 0.30, p<0.000). The mean pointing errors for the egocentric reference frame users revealed

that both the two reference frame users were easily to overestimate the homing direction when the turn angles were small (30°) while they were tend to underestimate the homing direction when the turn angles were larger (60° and 90°, Fig. 4-2B).

Figure 4-2: effects of navigation strategy and the turning degree on the behavioral performances which evaluated by the absolute error (A) or error (B) of the homing direction. Note: all the mean absolute errors were ranged from 5 to 15 degrees. A two way ANOVA test showed significant differences in the absolute error on factors of navigation strategies and turning eccentricity. The mean absolute errors indicate that subjects were overestimated the homing direction when the turn angles were small while the subjects were tended to underestimate the homing direction when the turn angles were larger.

4.2 Independent Component (IC) Clustering and Source

Localization

Allocentric Egocentric

Figure 4-3: The grand mean of the scalp map and individual scalp maps (A, B) as well as their equivalent dipole locations (C) for the parietal IC clusters for the allocentric and egocentric subjects. A total of nine from 10 allocentric subjects exhibit parietal components, and 7 of 7 egocentric subjects exhibited parietal component.

Allocentric Egocentric

Figure 4-4: The grand mean of the scalp map and individual scalp maps (A, B) as well as their equivalent dipole locations (C) for the occipital IC clusters for the allocentric and egocentric subjects. Panels as Fig. 4-2. 90.0% (9/10) allocentric and 85.7% (6/7) egocentric subjects have occipital ICs.

Allocentric Egocentric

Figure 4-5: The grand mean of the scalp map and individual scalp maps (A, B) as well as their equivalent dipole locations (C) for the left somatomotor IC clusters for the allocentric and egocentric subjects. Panels as Fig. 4-2. 70.0% (7/10) allocentric and 71.4% (5/7) egocentric subjects are with the left somatomotor ICs

Results of the component clusters showed that the spatial navigation at least involved several brain regions. The number of localizable brain ICs in allocentric and egocentric reference frame subjects did not differ significantly [F(1,2) = 0.05, p>0.05, n.s.]. Neither did the numbers of ICs contributed by allocentric and egocentric reference frame subjects to the 3 clusters differ significantly [group by cluster interaction F(2,2) = 1; p>0.05, n.s.]. Fig. 4-3, 4-4 and 4-5 shows the grand mean of the scalp map, the scalp maps from individual subjects and their equivalent dipole

models for three major IC clusters (parietal, left somatomotor and occipital clusters) across 10 allocentric and 7 egocentric reference frame subjects. The residual variances and Talairach coordinates of the equivalent dipole sources and the numbers of independent components were summarized in table-1 and table-2, respectively.

Table-1: The residual variances and Talairach coordinates of the equivalent dipole sources

Component Residual variance (%)

Talairach coordinates Distance to cluster center

Table-2: The Number of Components in the three IC Clusters

Parietal Occipital Left Motor Total

Number of subjects

4.3 Brain Dynamics of the Allocentric representation 4.3.1 The Parietal Component

The Fig. 4-6 shows the grand mean of ERSP images of the parietal IC cluster across 9 subjects who used the allocentric reference frame in the experiment. Results revealed that the alpha band power and its first harmonics slightly attenuated while subjects passed through the curve, and the alpha band power was increased during and after homing direction selections. The above changes in alpha band power were

observed in all turning degree.

Figure 4-6: The grand mean of the ERSP images (lower panels) and the average component map (top panel) of the parietal cluster across 10 subjects used the allocentric reference frame. Epochs were aligned to the onset of entering the tunnel, and straight segment was selected as baseline. Solid pink lines: the onset of turn. Blue dash lines: the offset of the turn. Brown dash lines: the onset of exiting the tunnel.

Black dash lines: the onset of selecting the homing direction. Red dash lines: the onset of selecting the homing angle. Note: the ERSP images showed that the alpha band power and its harmonics slightly decrease while passing through the curve, and then the alpha band power increases during and after selecting the homing angle. The above phenomena are observed in three cases which are with 30°, 60° and 90° turning angles respectively.

4.3.2 The Occipital Component

Figure 4-7 shows the grand mean of ERSP images of occipital IC cluster across 9 allocentric reference frame subjects. Results showed that the alpha band power slightly decreased when subjects passing through the tunnel turn. The alpha band power strongly increased before the end of turn and the increased alpha activity was sustained through the second straight segment and the homing directional selection.

When subjects responded to the homing angle selection, the stronger and sustained power decrease in alpha band was appeared and followed by another alpha band increase beginning around 1 sec after the end of the homing direction selection. No apparent differences on the EEG dynamics were found in the occipital IC cluster among three cases, which were composed trials with different turning angles.

Figure 4-7: The grand mean of the ERSP images (lower panels) and the average component map (top panel) of the occipital cluster across allocentric subjects. Panels as Fig. 4-6. Note: the alpha band power slightly decreases when subjects passing through the curve segment while it increases strongly at the end of tunnel. Then, the alpha power suppresses strongly during the homing angle selection. Finally, the alpha power increases again after the end of the angle selection. The above phenomena are observed in all cases.

4.3.2 The Left Motor Component

To characterize the changes of the neural activities related to process homing angle decisions at the somatomotor IC cluster, all trials were aligned to the onset of the homing direction selection and shows in Fig. 4-8. A strong alpha/mu activity near 10 and 20 Hz suppressed from the end of the tunnel turn and sustained to the end of the homing directional selection. Following the alpha blocking, an increase in alpha

band activity showed after the end of the homing direction selection. In comparison to the cases with larger turning angles (60° and 90°), a power increases in the beta band (roughly around the 20 Hz) clear presented in the case with the small turning angle (30°) after the end of the homing direction selection. No additional apparent effects of the turning angle were found in the ERSP images of the IC clusters.

Figure 4-8: The grand mean of the ERSP images (lower panels) and the average component map (top panel) of the left somatomotor IC cluster across 7 allocentric subjects. All the traces were aligned to the onset of the homing direction selection.

Panels as Fig. 4-6. Note: Strong alpha/mu activity was suppressed from the turn segment to the homing direction selection period. An increase in alpha activity was found after the end of homing direction selection.

4.4

Brain Dynamics of the Egocentric reference frame subjects

4.4.1 The Parietal IC clusters

The activities in alpha band and its harmonics were minor attenuated from around 1 to 3 sec after the entrance of the tunnel turn, and the higher alpha band power (around 10-12 Hz) sustain increased during the second straight segment and the period for determining the homing directions. After the end of the homing direction selection, the slightly increases in the beta band (around 14-30 Hz) apparently showed in the trails with 30° tunnel turns.

Figure 4-9: The grand mean of the ERSP images (lower panels) and the average component map (top panel) of the parietal cluster across egocentric subjects. Panels as Fig. 4-6. Note: Alpha band power and its harmonics attenuated while passing through the curve, and the alpha band power is increased while and after pointing the angle.

The larger tunnel degree revealed stronger alpha attenuation while passing through the curve but weaker wide band enhancement after button press.

4.4.2 The Occipital IC cluster

In the occipital IC cluster, strong increases in alpha band and its harmonics revealed 2 sec before the end of the turn segment and through the second straight segment and the power increase downward shifted to around 4-8 Hz during the homing directional selection (Fig. 4-10). Following the onset of homing direction selection, the alpha band and its first harmonics powers decreased again and sustained

to the end of the button press. A short duration of power increases near 10 and 20 Hz showed around 2.5 sec after the button press.

Figure 4-10: The grand mean of the ERSP images (lower panels) and the average component map (top panel) of the occipital cluster across egocentric subjects. Panels as Fig. 4-6. Note: We found strong alpha band power increased in the end of tunnel and alpha attenuation while subjects were pointing the homing angle consists in all tunnel cases.

4.4.3 The Left somatomotor IC cluster

The alpha/mu suppression showed from passing though the second straight segment of the tunnel and the homing degree and direction selection (Fig. 4-11). The clear power increases in the alpha/mu rhythms were only found in the trials with 60°

turn angles. The observed phenomena were similar to those found in the allocentric

reference frame subjects. Science phenomena of the right somatomotor IC cluster was similar with left somatomotor IC cluster, therefore we showed left somatomotor IC cluster only.

Figure 4-11: The grand mean of the ERSP images (lower panels) and the average component map (top panel) of the occipital cluster across egocentric subjects. Panels as Fig. 4-6. Note: Strong mu rhythm suppression showed from passing though the second straight segment of the tunnel and the homing directional and direction selection, then the increase activity in alpha band revealed around 1 sec after button press.

4.5 EEG differences between two strategy groups

To study the possible group differences in EEG activities more closely, we assessed the tonic power changes in EEG spectra of the three major IC clusters

between the two groups and the significance of those differences between the two groups were further confirmed by repeated-measures ANOVA.

Figure 4-12 shows the tonic and phasic changes in power spectra of the parietal IC clusters for two groups. The peak of the baseline power spectrum of the allocentric reference subjects was around 10 Hz while egocentric reference frame subjects exhibited the peak around 12 Hz. In comparison with the egocentric reference frame subjects, the baseline power of the subjects using the allocentric reference frames, were significant higher at the frequencies around 4Hz, 8-12Hz, 16-20 Hz and 36-40Hz. The above differences between these two strategy groups were not affected by the turning degree. The similar differences on the baseline power spectra between the two groups also revealed on the distribution of their phasic power spectra, but such differences did not reach the statistic significance. The occipital IC cluster also showed the significance difference on the peak of the baseline power spectra between the two types of subjects (Fig. 4-13). The baseline powers were significant higher at the frequencies at 8-11 Hz.

Figure 4-12: The averaged power spectral changes in tonic and phasic responses in parietal component. Turning angles are degrees of 30 (A), 60 (B), and 90 (C) respectively. Blue and red lines: mean spectral power of baselines for allocentric and egocentric subjects respectively. Light blue shadow marks the maximal and minimal of the phasic changes in the power spectra for the allocentric subjects. Light red shadow marks the maximal and minimal of the phasic changes in the power spectra for the egocentric subjects. Green horizontal lines mark the frequency ranges where the tonic power spectra are significantly different (p < 0.05) between allocentric and egocentric subjects. Note: for the allocentric subjects, the peak of the spectra is located at the low alpha band while it is located in the high alpha band for the egocentric subjects.

Figure 4-13: The averaged power spectra of the tonic and phasic responses in occipital IC clusters for allocentric and egocentric reference frame users. Panels as Fig. 4-12. Note: the peaks of the spectra are located near 10 Hz for both allocentric and egocentric subjects. Comparing with the egocentric reference frame subjects, the tonic power of subjects using the allocentric reference frame were significantly higher at the frequencies around 10-11 Hz (p < 0.05, mark in green horizontal line).

To further assess the relationship between the changes of the brain activities and the navigation performances, we further grouped the trials by their navigation performance and classified as the good estimation, under estimation and over estimation groups and redrew the ERSP images of these three groups. Results demonstrated that the navigation performance related changes on brain activities were in the parietal IC cluster of the egocentric reference frame subjects (see Fig. 4-14).

Comparing to trails with well-estimation of homing angle, the power attenuation at

the frequencies at 8-30 Hz (around alpha and beta band) was stronger when subjects overestimated homing directions, but the attenuation power was decreased when subjects were underestimated the homing directions. The Fig. 4-15 shows the quantitative comparisons of the performance related variations on parietal IC cluster

在文檔中探討環境參照系與自我參照系於空間巡航中腦波反應之差異 (頁 27-0)