Task mode Settings

The task mode setup includes the order of task modes and durations, and the time to switch maze. The modes of task contain searching task, motion condition, navigation followed by defined sign, fixation marker in the center of screen, all black display, one full screen picture, and a center fixation marker with one picture background. Except the searching task and navigation followed by defined sign, the other mode screen was halted at the last position. During the motion condition, four white rectangles randomly display on the upper, bottom, left and right sides of screen then participants must respond the correct direction by control keys.

2.1.3.3. Input/Output Settings

The output signal is an 8 bits event and grid position signal transmitted from parallel port (2pin ~ 9pin) and the input signal is a one bit signal received from parallel port for synchronizing the tasks switch. The input/output setup includes the parallel port number, whether enabling the input or output port or not, and the pin number of input port.

2.2. Questionnaire

The questionnaire was a technique commonly used by researchers to realize the feelings of subjects who participated in the behavioral experiment. After completing the experiment, subjects were asked to fill the questionnaire associated with searching

style. Subjects might make use of different strategies to search the target in the different mazes. The information of the questionnaire could let us know that the characters related to different typical landmarks were gave rise to the different impacts for subjects.

2.3. Subjects

2.3.1. Behavior Experiment Subjects

Thirty right-handed healthy adults gave informed consent before the study approved by local Institutional Review Board. Fifteen males and females participated in the experiment. Mean age of the females was 20.5 ± 3.5years and mean age of the males was 22.3 ± 4.6 years. None had a history of neurological or psychiatric disorders or any sign of color blindness or visual field defects.All participants were rewarded with NT$ 200.

2.3.2. EEG Experiment Subjects

Thirteen right-handed healthy females gave informed consent before the study before the study approved by local Institutional Review Board. Mean age of the females was 21.5 ± 3.6years. None had a history of neurological or psychiatric disorders or any sign of color blindness or visual field defects. All participants were rewarded with NT$ 800.

2.4. Behavior Data Analysis

2.4.1. Corrected Travel Distance of Path Ratio Analysis

The travel distance was the length of covered path when the participant navigated from the start position to the target position in each trial, and the corrected

travel distance was the ratio of the real path length which the participants travelled during a trial to the shortest path length (Figure 2-4). Path ratio was used to determine the performance of the spatial navigation. If subjects were familiar with the virtual environment, the value of path ratio would decrease.

Figure 2-4: The method of calculating the path.

2.4.2 Analysis of Variance (ANOVA) Analysis

The analysis of variance (ANOVA) simultaneously tests if the means of the populations are the same in the same significant level [31]-[34]. There are several assumptions about ANOVA. (a) Variables must be normally distributed. (b) Samples are independent. (c) Variances of populations are equal. The ANOVA estimates the variance within the populations and the variance of inter-populations by sampling data. The freedom of each population is used to apply the F distribution test.

The one-way analysis of variance (one-way ANOVA) is a technique used to compare means of two or more samples. The one-way ANOVA is used to test the average of groups and one-way ANOVA is used to test whether these differences are significant. It appears to test the average, but it analyzes the variance in fact. The one-way ANOVA is a robust procedure relative to violations of the normality assumptions. (a) Variables must be normally distributed. (b)The sample is a simple random sample (SRS). (c)Samples are independent. (d)Variances of populations are

equal. The ratio of variance is a comparison of the variance amongst the different

populations.

The one-way ANOVA measures significant effects of one factor only and the two-way ANOVA measures the effects of two factors simultaneously. The methods of ANOVA tests are divided into non-repeat experiment and repeated experiment. For non-repeat experiment of two-way ANOVA, the data are classified with two factors and the factors are tested individually to see whether there are significant impacts on the average for two factors. Classification variables are used to test the relationship between the independent variables and the dependent variables. We can see if there is any interaction between them. For repeated experiment, one-way ANOVA is used to test the different values of the same group in the different situations. For repeated experiment of two-way ANOVA, the above test is processed in addition and then we can observe the impact of the interaction between two independent variables and the dependent variable.

2.5. EEG Data Analysis

2.5.1. EEG Data Analysis Procedures

Fig. 2-5 showed the schema of the data analysis procedure for EEG signals. The EEG data were recorded with 16-bit quantization level at a sampling rate of 2000 Hz .The EEG signals were down-sampled to sampling rate from 2000 to 500 Hz. The EEG signals were filtered with a low-pass with the cut-off frequencies at 50 Hz and then the EEG signals were filtered with high-pass filter with the cut-off frequencies at 0.5 Hz. The EEG data were extracted epochs from 1 s before the deviation onsets and 23 s after the deviation onsets. We would remove some epochs that were consisted too many noisy signals. These remainding epochs were combined with each other. The EEG data were processed by independent component analysis (ICA) and we applied

the method of Event-Related spectral perturbations (ERSP) to investigate the brain dynamics responses.

Figure 2-5: The schema for EEG analysis

The EEG responses related to different navigation stages of a block were extracted and combined together. There were several types of artifacts which were identified and eliminated from EEG data by using the EEGLAB toolbox. Criteria of artifact rejection included linear drift (Figure 2-6 (a)) and abnormally distributed data (Figure 2-6 (b)). The artifacts might be generated from equipments or the physiologic activity like as muscle artifact. Eye movements usually were observed in frontal electrodes. Sweat artifact was characterized by very low-frequency oscillations (Figure 2-6 (c)) [35]. The high impedance at posterior temporal electrode T6 resulted

in this electrode recording from the ground on the forehead (Figure 2-6 (d)) [35].

(a)

(b)

(c)

(d)

Figure 2-6: The artifacts were existed in the EEG signals. (a): Artifacts were generated from the bad channel T7. (b): Artifacts were muscle activities. (c) The distribution here (midtemporal electrode T3 and occipital electrode O1) were suggested sweat. Note that morphology and frequency were also consistent with slow rolling eye movements [35]. (d) The high impedance at posterior temporal electrode T6 results in this electrode recording from the ground on the forehead [35].

2.5.2. Independent Component Analysis (ICA)

The broad spread of EEG source potentials through the brain, skull, and scalp at each scalp electrode. The ICA algorithm was suitable to decompose the EEG data which was collected with single scalp channels. The ICA algorithm was related to the method called blind signals separation and it was a powerful approach to identify the complex spatiotemporal dynamic [36]-[37].The ICA was used to find a linear signal separation and feature extraction. The ICA performed a linear un-mixing of multi-channel EEG recordings into temporally independent statistical source signals.

After ICA decomposition, the spatial filters were chosen the maximum among temporally independent signals in the mixed channel data. These information sources might represent synchronous activity. The application of the ICA algorithm was solved identification, localization and separation [38]-[40]. (1) The component source locations to the scalp sensors were fixed throughout the data. (2) The source activities of mixing source activities were summed linearly and instantaneously. (3) The component source activity waveforms were temporally independent of one another. (4) There were no differential delays involved in projecting the source signals to the different sensors. (5) The probable distributions of the individual component source generated by nonlinear cortical dynamics activity values were not precise Gaussian.

(6)The signal source of physiological activity was not time locked to the sources of the EEG data. We would adopt independent component analysis (ICA) to find spatial filters for information sources of the EEG data. Figure 2-7 showed the process about ICA decomposition data which were transform into temporally independent processes.

ICA applied to a matrix of EEG scalp data to find a non-mixed matrix of weights and non-mixed matrix were permuted by weights.

The sources of EEG activities were regarded as reflecting synaptic activity of cortical neurons. An independent component of the EEG data composed of the fixed

scalp map and a time series. The fixed scalp map showed the relative weights of the projection from each electrode location. The information of time series was relative to the amplitude and the polarity at each time point. In matrix algebra form, the scalp data were X, the component activations were U, and the ICA model estimated a linear mapping W such that the unmixed signals U were mutually independent.

X

U =W . (2-1)

We would apply spatial filters W to the EEG data to derive activation time courses of the independent component processes. The reconstituted data from the components were used by conversing process. were the component mixing matrix.

−1

W

U

X = W⁻¹ (2-2)

Figure 2-7: Schema for Independent Component Analysis (ICA) data decomposition and back-projection [39].

Fig. 3-5 showed the result of the scalp topographies of ICA weighting matrix.

ICA weighting matrix projected each component onto the surface of the scalp. We could observe the artifacts and channel noises were effectively separated from the signals.

Fig. 2-8: These were the scalp topography of ICA decomposition. The color bar was the amplitude of component signals.

2.5.3. Component Clustering

Two equivalent EEG sources in different subjects might project to the same electrode location with variable amplitude [41].The component clustering of the cross-subject was classified several significant clusters of the brain activity. To cluster these components from different subjects was determined by the information about scale maps, spectra, event-related potentials, time-frequency results, dominant activity patterns, source localization information [42]. Independent component clustering was required to compare ICA decompositions from all subjects [39]. In the similar component clusters, we meant the values of these components to get individual component.

2.5.4. Event Related Spectral Perturbations (ERSP) Analysis

Event-related spectral perturbations analysis was time-locked but not phase-locked. ERSP was first proposed by Makeig [43]. An event-related spectrogram was averaged power or log power of frequency / time values, before applying Fast Fourier Transform with overlapped moving windows and we could choose the spectral baseline. We would normalize the result by subtracting the baseline period.

The procedure was applied to all epochs, and the color-coded image of mean log spectral differences were the ERSP image. The ERSP image showed spectral differences after events. Figure 2-9 showed the steps of the ERSP [44]. FFT was applied in each window with 256 samples.

Figure 2-9: The data was processed by using ERSP analysis [44].

2.5.5. Permutation Tests

The permutation test was a type of statistical significance test. The reference distribution was obtained by calculating all possible values of the test statistic under rearrangements of the data [45]. Permutation tests tried to find all possible combinations and permutations of the data. Permutation tests were suitable for the comparison of two populations with features were not the normal distribution or a small number of samples. There were two groups of global data and local data in our experiment. We adopted permutation tests [46]-[48] to estimate the probability distribution of these groups. The sampling distribution showed what would happen if we took many samples under the same conditions. Figure 3-6 showed the procedure of permutation test. We put together the ERSP data of different typical landmarks into two matrices form all subject. First, the matrix of global landmarks subtracted the matrix of local landmarks to get the new matrix. We processed the t-test for the new matrix. If the result was showed significance (p < 0.05), we continue to execute the second step. There were three steps in the second state. The elements of relative row values of the same subject between these two matrices were exchanged randomly with each other. Then, we processed t-test like as first state. We would repeat the second state 10000 times to see if the result was significant or not.

Figure 2-10: The procedure of permutation tests.

III. Spatial Navigation in Virtual Environments with

Different Types of Landmarks: Behavioral Experime nt

3.1. Methods of the Behavioral Experiment

Fig. 2-4(a) showed the experiment environment. Subjects looked at the screen which was displayed the virtual environment of the maze, and subjects navigated in the maze by using keyboard. The shielding room was used widely in various areas of researches to prevent foreign interference. Fig. 2-4(b) showed the shielding room which was used to prevent magnetism, sound, electro-magnetic wave, and electronic signals. The shielding room was consisted of single acoustic wall, single acoustic ceiling, and single floating floor. Subjects processed the experiment in the shielding room avoiding outside interference. Subjects could be monitored from the camera on the desk.

(a) (b)

Figure 3-1: (a)Experiment environment. (b)Shielding room.

3.1.2. Behavioral Experiment Procedure

All participants were tested individually by similar global or local landmarks in the virtual maze environments. In the beginning, subjects were unfamiliar with the virtual environments. For this reason, we designed the training state in order to help subjects to familiar with environments of different typical landmarks. Subjects could adopt different typical landmarks to navigate the virtual environment in the training state. Figure 3-2(a) showed the order of behavior experiment in training state and testing state. A trial was composed of 12 blocks at most and subjects searched only one target in a block. Subjects searched totally 12 targets at most in the single type landmarks environment. Subjects searched each target without time limit until all 12 targets were found unless up to 6-minutes limit. Subjects learned the positions and locations of targets in the training state. Then the participant searched each target within 20-seconds time limit each target in the testing state. Only one target appeared in a task and the other one would not appear until the current target was found or the 20–seconds time limit which was reached. Although the order of the targets order was randomly assigned, there were the same times of targets. After finding the target, the scene would be transferred randomly to other position and a new target picture was appeared to start next searching task. There were two sequences of landmarks when subjects participated in the spatial navigation experiment. One sequence consisted of eight consecutive trials of global landmarks followed by another eight consecutive trials of local landmarks, and the other consisted o eight consecutive trials of local landmarks followed by another eight consecutive trials of global landmarks. Subjects were randomly assigned to one sequence to avoid confounding from order effect

caused by a particular sequence.

Figure 3-2: Schema of behavior experiments with different landmarks.

3.2. Results

3.2.1. Training Trials Associated with Gender and Landmarks

The main effect of landmark was not significant (F (1, 28) =.595, p > .05, η²=0.021, Figure 3-3 (a)). The participants needed equivalent number of learning trials to reach the criterion of 100% performance (Global: 3.43±0.37 trials; Local:

3.83±0.39 trials, Figure 3-3 (b)). The main effect of gender was also significant (F (1, 28) = 9.946, p < .01, η²=0.262). In general, the male participants needed fewer learning trials (2.77±0.39) than females (4.5±0.39) to fulfill the criterion success rate.

None of the interactions reached significance (all ps > .05). Males had better performance to search targets than females. Males needed fewer trials to reach 100%

success rate than females. Although time was restricted within 6-minutes in the training state, the training state was effective for males. Males achieved 100% success rate faster. The results indicated that males became familiar with the virtual environment earlier than females did.

(a) (b)

Figure 3-3: (a) Different landmarks needed different trials to reach 100% success rate.(b) Males and females needed different trials to reach 100% success rate

3.2.2. Travel Duration

The travel duration was the amount of time passed between the time when participants started out to find a given target as instructed and the time when they reached the target location. There was no significant main effect of landmark (F (1, 28)

= 3.133, p > .05, η²=0.101, Figure 3-4 (a) ), but there were significant main effects of trial (F (7, 196) = 20.342, p< .01, η²=0.421, Figure 3-4 (b) ) and gender (F (1, 28) = 19.128, p< .01, η²=0.406). In general, the participants spent less time to find the targets in the later trials (the last trial: 8.32±0.21 seconds) than in the earlier trials (the first trial: 11.10 ± 0.29 seconds), and the male participants needed fewer time than female participants. There were also a significant interaction between landmark and gender (F (1, 28) = 6.79, p< .05, η²=0.195), and between landmark types and trial (F (7,196) = 2.611, p< .05, η²=0.085). The curves of global and local landmarks were implied that the performance of subjects was got better and better in the later trials.

The phenomenon indicated that subjects became more familiar with the virtual environment so that subjects spent little time searching targets. Subjects had similar performance in the virtual environment with global or local landmarks in the whole

trials and then we analyzed travel duration of time about gender. All participants got better and better manifestation of searching time across the experiments. Males wasted little time searching targets in both landmarks environments. Males had better cognitive performance of searching time than females regardless of different landmarks in every trial. Males and females improved their performance in the later trials.

(a) (b)

Figure 3-4: (a) Subjects wasted time searching the target in the virtual reality environment with different landmarks in each trial. (b)Searching time of males and females in each trial. Males spent less time than females for each trial. Both males and females’ travel duration get shorter with the number of trials performed.

The analysis on the simple main effect revealed that male participants spent similar amount of times before finding the target position in both types of landmark’s environments ( Figure 3-5 (a)), while the females spent less amount of time to finish a trial in the local than in the global condition( Figure 3-5 (b)). Furthermore, the participants had unequal improvement in local and global landmark environments.

Neither the trial × gender nor the three-way interactions reached significance (trial × gender: F (7, 196) = .482, p >.05; three-way interaction: F (7,196) = .365, p >.05).

(a) (b)

Figure 3-5: (a) Male wasted time navigating in the mazes with different landmarks.

Males had the similar performance of global and local condition. Males improved his performance in the later trials. (b) Female wasted time navigating in the mazes with different landmarks. Females consumed the less time in the local condition in the whole.

3.2.3. Path Ratio Results

The main effect of landmark was significant (F (1, 28) = 56.946, p< .01, η²=0.670, (Figure 3-6)). The corrected travel distance was longer for the local condition (1.72±0.05) than for the global condition (1.46±0.03). The main effect of trial order was also significant (F (7,196) = 14.962, p < .01, η²=0.348). The three-way interaction was significant (F (7,196) = 2.953, p < .01, η²=0.095). The global condition had better performance than the local condition for all subjects in comparison of path ratio.

Figure 3-6: The different path ratio values were between global landmark environments and local landmark environments in each trial. In the whole, subjects covered shorter walking path in the global than in the local condition. There was the variation in the local condition in comparison of path ratio.

In the global condition (Figure 3-7(a)), the females had longer corrected travel distance than males in the first trial, while they had equivalent corrected path length in

In the global condition (Figure 3-7(a)), the females had longer corrected travel distance than males in the first trial, while they had equivalent corrected path length in