On-line implementation of the Subband-AR-EEG Viewer

Due to its simplicity, the algorithms provides a robust tool for on-line processing re-quired for meditation EEG interpretation as well as the biofeedback scheme in the BCI (brain-computer interface) research. Currently, we have implemented the MEEGI algorithm under Simulink (MathWorks, Inc., Natick, MA) with Real-Time Workshop on a Pentium-M 1.4 (GHz) notebook (Fig. 2-14(a)). By generating a real-time code with Real-Time Workshop, the algorithm can be downloaded to the kernel and run in a real-time manner under Windows [Guger 2001]. The classification results were displayed on a monitor. As shown in Fig.

2-14(b), the height of each bar reflects the power percentage of the corresponding EEG rhythm within a 2-sec frame. We can thus monitor the meditator’s state in a real-time manner that enables the development of a more subtle correlation between the EEG characteristics

and the meditation scenario.

χ δ θ α β

300 sec

600 sec

900 sec

1200 sec

(a)

χ δ θ α β

300 sec

600 sec

900 sec

1200 sec

(b)

χ δ θ α β

300 sec

600 sec

(c)

Figure 2-13: Running gray-scale charts (Channel F3) for two meditators (a) subject 2k1019p, and (b) subject 2k0830a, and (c) one non-meditator (control) subject

(a)

(b)

Figure 2-14: On-line implementation of the Subband-AR-EEG Viewer. (a) Computer 1 executes the MEEGI algorithm (Fig. 2-10), and Computer 2 displays the classification results, as illustrated in (b). The height of each bar reflects the power percentage of the corresponding EEG rhythm within a 2-sec frame.

Chapter 3－

Investigation of Visual Perception under Zen-Meditation

If the brain were so simple we could understand it, we would be so simple we couldn't.

~Lyall Watson

ne topic of interest in the meditation study is the evoked potentials (EP) (or event-related potentials, ERP) of a practitioner under meditation, which includes auditory evoked potential (AEP), somatosensory evoked potential (SEP), visual evoked potential (VEP), and so on. Each parameter is meaningful to the respective perception function. Due to unusual perceptions often experienced during meditation, human brain in response to external flash stimuli during Zen meditation drew our attention. Recording of VEPs provides a means of characterizing the visual pathway and visual function. VEPs can be recorded by applying either patterned or non-patterned stimulus that results in various VEP waveforms [Odom et al. 2004]. Since practitioners must close their eyes during meditation, we employed non-patterned flashes in the VEP recording.

O

3.1. Why EEG-triggered F-VEP?

In Zen-meditation EEG study, increased alpha activity over the frontal regions of the

and so has the increased frontal alpha coherence [Murata 2004]. Kasamatsu and Hirai found an increase in alpha amplitude at the beginning of meditation, which then spread frontally [Kasamatsu and Hirai 1966]. Furthermore, Takahashi et al. observed that the increased frontal alpha power correlated with the enhancing internalized attention [Takahashi et al. 2005]. Thus increased frontal alpha activity was hypothesized as a result of Zen-meditation process. All the observations have led into further understanding of the function-correlated, spatial char-acteristics of the brain affected by meditation.

In another aspect, Zhang et al. [Zhang et al. 1993] claimed that the amplitudes of F-VEPs (VEPs under flash stimuli) of Qigong meditators increased under meditation. In Xu et al.’s research [Xu et al. 1998], the amplitude of F-VEP increased while the latency decreased. They suggested that concentration and attention may be the reason of altering the evoked potentials.

One problem encountered in the F-VEP study is to determine the appropriate timing for ap-plying the flash-light stimulus. To our knowledge it has not been reported in regard to this is-sue. In our previous study, stimulus was applied at the mid-section of meditation at which subjects might undergo various physiological and mental states. In that case, F-VEPs were not able to reveal different experimental courses such as the section before, during, and after meditation [Liu and Lo 2005]. To assure that all F-VEPs are acquired under a consistent con-dition, a rational experimental setup is to apply the stimulus based on a controllable factor. To gain access to particular brain states, one approach is to ask the subject to signal the attain-ment of the meditation state by finger moveattain-ment [Newberg et al. 2001, Lo et al. 2003, Taka-hashi et al.2005]. However, it often causes meditators to break off from the meditation state.

To investigate the ERP activities in a given brain state defined by EEG, we thus conceive the idea of EEG-triggered F-VEP scheme, that is, the flash-light stimulus is applied under specific oscillatory features of the EEG. In this preliminary study, we intuitively selected the frontal α-rhythm as the F-VEP triggered signal based on the results reviewed previously and our empirical observations these years. The following section illustrates the methods for

de-tecting the frontal α-rhythm and the experimental setup. Significant results obtained are pre-sented at the end of this chapter.

3.2. Alpha-dependent F-VEP

According to the description in 3.1, the flash-light stimulus is to be applied upon emer-gence of the frontal α-rhythm. The scheme, Subband-AR EEG Viewer, proposed in Chapter 2 [Liao and Lo 2006] provides more accurate estimate with better resolution [Hayes 1996, Güler et al. 2001] and, in particular, allows on-line α-rhythm detection within a very small time frame. Modification of the scheme for on-line α-rhythm detection is described below.

3.2.1. Online α-rhythm detection

For online α-rhythm detection, the SARD algorithm described in Chapter 2 was slightly modified. Because cutoff frequency of the bandpass filter is different (to be explained in next section) for VEP recording, the tree structural filter bank in Fig. 2-5 was adjusted as shown in Fig. 3-1. As presented in Chapter 2, an AR(2) model can be expressed as

]

The model coefficients can be determined by solving the autocorrelation normal equations [Hayes 1996]. After the model coefficients have been obtained, the conjugated pole pair is determined as,

]

Thus the root frequency of the signal can be obtained from Eq. (3-2)

⎟⎟

erated to attain good accuracy in discriminating between α and δ/θ rhythms. The equivalent cutoff frequency is 15Hz. We designed a criterion based on the root frequency to detect the α-rhythm. The algorithm examines each windowed segment to check whether it meets the following criterion.

Criterion-α: f_r,1<14Hz and 7Hz<fr,2.

The root frequency fr,1 is used to differentiate EEG rhythms in 0–14Hz from those in 14–30Hz.

Next, root frequency fr,2 is examined to screen out δ/θ rhythms.

Note that output1 and output2 are the results of downsampling (Fig. 3-1), the root fre-quency fr,i should be further divided by 2ⁱ. This experiment employed a window length of 1 second, with a moving step of 0.5 second.

Figure 3-1: Tree structural filter bank for the Subband-AR EEG Classifier. H(z) is de-signed by least-squares error minimization with cutoff frequency 30Hz.

3.2.2. Simulation

To verify the effectiveness of the α-rhythm detection algorithm, the algorithm was firstly applied to a simulated signal. We assumed a sampling rate of 128Hz. The signal can be simu-lated by the pole placement method, that is, by placing each pole in the corresponding fre-quency band (Table 3-1) and adding Gaussian noise.As displayed in Fig. 3-2(e), the simu-lated 10-second signal was formed by connecting four segments of δ, θ, α, and β-rhythm

pat-LPF

Signal Compute

root

Classificatio n algor ithm

terns. Detection result in Fig. 3-2(e) is illustrated by two gray scales, with dark (light) gray indicating the α (non-α) pattern. The result clearly justified the effectiveness of the algorithm in α detection.

Table 3-1: Locations of poles of the simulated signal

δ θ α β

Poles’ location 0.98∠0.04 0.98∠0.16 0.98∠0.4 0.88∠0.63

δ (a)

θ

(b)

β

(c)

α

(d)

(e) non α

α

1 sec

Figure 3-2: Classification result of the simulated signal. Different grays are used to il-lustrated the α and non-α patterns.

3.2.3. Off-line alpha detection

Empirical EEGs often exhibit highly complex, irregular rhythmic patterns that make the recognition of specific EEG pattern more difficult. Our algorithm is robust in dealing with this

kind of complication. Figure 3-3 displays the result of α detection. The error rate, estimated from the results of identifying 780 alpha candidates, was approximately 7.2% (4.6% false negative and 2.6% false positive rate) in comparison with the results of naked-eye examina-tion by an experienced EEG interpreter.

-200 0 200

μV

-200 0 200

μV

-200 0 200

μV

-200 0 200

μV

non α

α ^{1 sec}

Figure 3-3: Result of α detection for real EEG signal.

3.2.4. F-VEP

The F-VEP consists of a series of negative and positive peaks, denoted respectively by N and P followed by a number. The number is referred to as the order (or time) of occurrence of that particular peak from the stimulus. The F-VEP source is located in the occipital lobe.

Normally, the event-related brain potentials propagate via neural network toward the nearby regions. The phase differences among different channels are caused by the time delays of

brain-wave propagation [Hughes et al. 1992]. Figure 3-4 shows typical F-VEPs recorded on Fz, Cz, and Oz with corresponding peak labels. In this study, peaks of significance including N2, P2, N3, and P3 are to be analyzed. Note that F-VEPs of different channels have phase de-viation.

Researchers inferred that the noticeable negative peak N2 of Oz was generated in lamina IV cb [Kraut 1985, Ducati 1988], then the following positive peak P2 might reflect the inhibi-tion activity within lamina. This study is mainly based on the hypothesis that Zen meditainhibi-tion affects visual neural pathway that can be revealed on the F-VEPs.

3.3. Experimental setup and protocol

3.3.1. Subjects

This study involved 11 meditators and 11 control subjects. In the experimental group, 4 females and 7 males at the mean age of 27.5± 3.2 years participated. Their experiences in Zen-Buddhist practice span 5.5 4.3 years. The control group included 3 female and 8 male students with an average age of 23.6

±

± 3.3 years.

3.3.2. Apparatus

The EEG signals and F-VEPs were recorded at standard 10/20 positions (Fig. 3-5) with 32-channel SynAmps amplifiers (manufactured by NeuroScan, Inc.) connected to a Pentium-4 (1.5 GHz) PC. Common reference of linked M1-M2 (mastoid electrodes) was used. EEG sig-nals, after amplification, were pre-filtered by a bandpass filter with passband 0.3–50 Hz, and digitized at 1000 Hz sampling rate. A 60-Hz digital notch filter was applied to the data to re-move artifacts from power line or the surroundings.

0 50 100 150 200 250 300 350 400 450

Figure 3-4: Profile of F-VEPs on (a) Fz, (b) Cz, and (c) Oz with corresponding peaks labeled.

Figure 3-5: 32-channel recording montage.

We developed an online α-detection algorithm that was implemented by using g.BSamp with g.RTsys (manufactured by Guger Technologies, Inc.) connected to a Pentium-M (1.4 GHz) notebook. g.BSamp is a stand-alone biosignal amplifier and g.RTsys is a biosignal ac-quisition and real-time analysis system for notebook implementation. To facilitate the real-time α detection, channel-Fz EEG was pre-filtered by a bandpass filter with passband 0.5–30 Hz, and digitized at a lower rate of 128 Hz. The α-detection algorithm was imple-mented on Simulink (MathWorks, Inc., Natick, MA) with Real-Time Workshop. By generat-ing real-time code with Real-Time Workshop, the algorithm can be downloaded to the kernel and run in a real-time manner under Windows [Guger et al. 2001]. The experimental setup is shown in Fig. 3-6. The subject stayed in an isolated space. A CCD camera positioned in front of the subject was used to monitor the entire procedure. The 32-channel EEG signals were recorded by an EEG recording system. Another computer read channel Fz simultaneously to identify the occurrence of frontal-α rhythm. Once the frontal-α was ascertained, the computer

triggered the flash light controller to generate flash stimuli.

CCD camera α-rhythm

de-tection

EEG recording system trigger

Flash light

Flash light con-troller

EEG

Figure 3-6: Experimental setup for α-dependent F-VEP recording.

3.3.3. Experimental paradigms

During the experiment, subjects sat in a separated space in the laboratory. Each recording lasted for about 60 minutes, including three sections: two 10-minute background EEG re-cordings (section I and III) before and after a 40-minute recording (section II, the “main sec-tion”) of EEG under meditation (experimental subject) or rest (control subject). The control subject sat in a normal, relaxed position with eyes closed, while the meditators practiced Zen-Buddhist meditation during the 40-minute main section. In Zen meditation, the subject sat, with eyes closed, in the full-lotus or half-lotus position. Each hand formed a special mudra (called the Grand Harmony Mudra), laid on the lap of the same side. The subject fo-cused on the Zen Chakra and the Dharma Eye Chakra (also known as the “Third Eye Chakra”) in the beginning of meditation till transcending the physical and mental realm. The Zen Chakra locates inside the third ventricle, while the Dharma Eye Chakra locates at the hy-pophysis [Lo et al. 2003].

The term “alpha-dependent F-VEPs” was used because we recorded F-VEPs upon the detection of frontal α rhythms. One run of alpha-dependent F-VEPs were recorded in each of the three sections (Fig. 3-7). Each run consisted of 50 alpha-dependent flash stimuli. The

in-terval between two consecutive stimuli was longer than 1 sec. The flash light, with a 10μs du-ration, was produced by a xenon lamp that was placed 60 cm in front of the subjects’ eyes.

Alpha-dependent F-VEPs were acquired from midline channels Oz, Cz and Fz, with the linked-mastoid electrode as the reference. Since we employed the mastoid-referenced unipolar montage, alpha activities could be found in the frontal channels of all subjects [Niedermeyer and Lopes da Silva 2004]. However, more frontal alpha activities were detected during medi-tation, that reduced the time required for collecting 50 alpha-dependent F-VEPs.

Section I Section II (main section) Section III 50 alpha-dependent stimuli

25 30

10 50 min

0 60

Figure 3-7: Scheduling of the F-VEP recording procedure.

3.4. Modulation of F-VEP amplitudes due to Zen-Meditation process

As shown in Fig. 3-8, different codes were used to indicate the stimuli presented in dif-ferent sections. The stimuli in section I, II, and III are marked by code 128, 64, and 32, re-spectively. And the marker at the upper left of each code indicates the time of flash stimula-tion. A concluding F-VEP of each section was derived by averaging 50 raw tracings in one run.

09:38 09:40 09:42 09:44 09:46 Oz

Pz Cz Fz

128 128 128 128 128 128

Figure 3-8: Display format of selected channels (Fz, Cz, Pz, and Oz) for α-dependent F-VEP recording. The vertical bar at the upper left of mark ‘128’ indicates the time of applying flash stimulus.

Inter-subject variations of human VEP under the open experimental environment are complicated. We thus investigated the intra-subject differences among various sections con-ducted in one experiment. We measured the amplitude and latency of the average F-VEP, and quantified the difference between various sections. Our results presented quite different trends between two groups. Figure 3-9 plots an example of the F-VEPs at Fz, Cz and Oz (from the top) for one subject. The solid lines represent the F-VEPs in the section I, and the dash and dot ones stand for those in the section II and III.

0 100 200 300 400 500

Figure 3-9: The α-dependent F-VEPs of one meditator recorded on (a) Fz, (b) Cz, and

Table 3-2: The changes in the peak amplitudes of specific F-VEP components.

N2-P2 115.70 92.76

0.040*

88.77 110.03

0.052 0.0025* 0.045*

Fz

P2-N3 143.35 128.01 NS 99.77 119.34 NS NS NS

I: section I, II: section II (main section), III: section III,

*: P<0.05,

NS: Not Significant

From our results we found that latencies of all components exhibit no significant ence among all sections in both groups. However, the variations of amplitudes show differ-ences in some components between two groups. Table 3-2 presents, for each F-VEP compo-nent, ratios of the group average amplitudes of different sections. The p value is calculated by paired t-test (compared with selves in different phases) and t-test (compared to the other group in the same phase). Amplitudes of P1-N2 and N2-P2 on Cz and Fz increased signifi-cantly during meditation, yet, decreased during relaxation in the control group. On the other hand, N2-P2 amplitude on Fz decreased after meditation (experimental group) but increased after rest (control group). Apparently, the transit from one to another section caused F-VEP

amplitudes on Cz and Fz to vary in opposite directions for both groups. We also observed sig-nificant differences between two groups in P1-N2 and N2-P2 amplitudes (Oz). P1-N2 ampli-tude (Oz) decreased during meditation but increased during rest. N2-P2 ampliampli-tude increased in the control group, but had little change in the experimental group. Contrary to the earlier peaks P1-N2 and N2-P2, N3-P3 amplitude (Oz) in the control group slightly decreased, whereas this peak amplitude increased in the experimental group.

Among all the F-VEP components, the most noticeable difference between two groups was the N2-P2 of Cz and Fz. Figure 3-10 plots the variations of N2-P2 amplitudes at Cz and Fz. Each bar represents the percentage of F-VEP varying from section I to section II

⎟⎠

⎜ ⎞

⎝

⎛ − ×

% I 100

I

II for one subject (white: experimental subject, gray: control subject).

Signifi-cant distinction is observed between two groups. Most meditation practitioners had their N2-P2 amplitudes increasing by an average rate of 20.1% (standard deviation among 11 sub-jects was 15.12%). Control subsub-jects, on the contrary, exhibited a decreasing trend (average rate: −6.36%, standard deviation: 7.78%).

Fz

The ratio of increment (%)

(a): Fz

The ratio of increment (%)

(b): Cz

Figure 3-10: Variations of N2-P2 amplitudes at (a) Fz and (b) Cz. Each bar represents the percentage of F-VEP varying from section I to section II ⎟

⎠

individual subject (white: experimental subject, gray: control subject).

Chapter 4－

Investigation on Spatiotemporal Characteristics of Zen-Meditation EEG Rhythms

Mathematics compares the most diverse phenomena and discovers the secret analogies that unite them.

~ Joseph Fourier

he work presented in this chapter goes further with a focus on the multi-channel meditation EEG analysis to explore the spatial-spectral behavior of Zen tion EEG. We will report the findings of the spatiotemporal behaviors of medita-tion-EEG spectra based on the analyzed results of 30-channel EEGs. To deal with such an enormous amount of EEG data, we developed a novel system, meditation EEG interpreter, that was modified from the MEEGI algorithm described in Chapter 2. This interpreter is ca-pable of identifying various EEG activities and detecting the artifacts.

T

4.1. The meditation EEG interpreter

Artefacts occur every now and then during EEG recording. To identify artifact-interfered EEG, the MEEGI algorithm was further enhanced to facilitate the long-term meditation EEG analysis and to eventually provide an overview of the entire meditation EEG record.

4.1.1. The algorithm

The meditation EEG interpreter is able to identify six wave patterns and two artifacts of-ten appearing in meditation EEG. The six wave patterns include: (1) low-power, almost flat wave (denoted by ‘

φ

’), (2) multi-frequency activities with approximate power (denoted by

‘χ’), and (3) δ, (4) θ, (5) α, and (6) β activities. Wave pattern

φ

refers to the EEG activity with amplitude no larger than 20

μ

V. Wave pattern χ mostly appears at the transition from one EEG rhythm to another, accordingly, no dominant rhythmic pattern can be justified for the segment.

In addition to the six patterns above, the interpreter can detect two artifacts: baseline-drift (denoted by ‘B’) and EMG (electromyograph) interference (denoted by ‘I’).

Each windowed segment of the EEG is examined by the following criteria in order (Fig.

4-1):

Criterion-

φ

: Amplitude<20

μ

V,

Criterion-χ: fr,3<7Hz<fr,1 and ⏐p3⏐>0.8, Criterion-δ: fr,1<7Hz and fr,4<3.5Hz, Criterion-θ: fr,1<7Hz,

Criterion-α: 7Hz<fr,1<14Hz and 7Hz<fr,3, Criterion-β: fr,1>7Hz,

where p3 is the AR(2) pole of output3. The length of p3 is closely related to the amplitude and reflects the significance of the root frequency fr,3. In the criterion, the major root fre-quency of each segment (fr,1) is used to differentiate between 0–7Hz and 7–30Hz EEG rhythms. A segment is identified as the χ activity if its fr,1 is higher than 7 Hz and its fr,3 is lower than 7Hz with ⏐p3⏐ larger than a given threshold. Then the fr,4 is used to discriminate δ activity from θ activity. The pattern excluded by the α-criterion, yet, with fr,1 higher than 7Hz is classified as β activity. EEG segments with the above characteristics exhibit no particularly dominant EEG rhythm. According to the oscillations assumption for the brain dynamics,

dif-ferent neuronal networks may start to oscillate with difdif-ferent frequencies during mental activ-ity [Klimesch 1996, Pfurtscheller and Lopes da Silva 1999]. It is thus reasonable to recognize this as the β activity.

After identifying the six patterns from the meditation EEG record, the algorithm further scrutinizes the interpretation to detect the segments of baseline drift and EMG interference possibly leaked to the categories of normal EEG patterns.

4.1.2. Baseline drift detection

EEG signals very often are affected by baseline drift caused by eye movement, breathing, etc. Some methods have been proposed to remove this baseline drift [Philips 1996, Lo and Leu 2001]. In this proposed algorithm, the baseline drift component is detected and marked without being removed from the record. Due to the low-frequency characteristic of baseline drift, the criterion for detecting baseline drift is focused on the delta band (1–3Hz) as de-scribed below.

(1) A segment of EEG which has been classified as δ and is longer than 1.25 s.

(2) Amplitude of the signal exceeding 80

μ

V.

(3) Number of zero-crossing events of the segment is less than 3 within 1 s.

(3) Number of zero-crossing events of the segment is less than 3 within 1 s.

在文檔中 以自迴歸模型為基礎之禪坐腦電波隨時空變異之頻譜分析 (頁 42-0)