Estimation of motor imaginary using fMRI experiment based EEG sensor location.
Keywords: BCI, EEG, fMRI, SMA, LDA
Brain computer interface (BCI) is a computer interface system, which is based on brain activity as determined by voluntary will. Using the BCI, we can control a computer system via imagination or other mental tasks. In our BCI research, we consider a BCI system, which is based on non-invasive EEG signals, and endogenous mental activity was used , . In order to improve the performance of the BCI system, it is important to get reliable EEG signals. In order to get reliable EEG signals, noise and artifacts from EEG signals need to be removed. Noise can be removed by anti-electromagnetic or by enhancing amplifier performance or by other hardware-type methods. Also, the removal of artifacts means the illumination of brain activity element from EEG signal except for the brain activity elements, which are related with the specific mental tasks of the BCI system. In order to remove the artifacts, we need to determine mental tasks carefully. If an inappropriate mental task is used (for example, mental tasks which induce little distinct change in EEG signals feature), it is difficult to get meaningful EEG signals. After the mental tasks are determined, we need to find the proper location of the EEG sensor according to the specific mental task. Brain activity, according to specific mental tasks, is considerably localized, and the location where we can expect the features of the EEG signal is probably also localized according to the specific mental tasks.
The conventional BCI researches are mainly focused on developing more plausible signal processing methods that can obtain more accurate EEG signals. These kinds of conventional BCI systems apply almost same approach for deciding mental tasks and locations of EEG sensors. Mental tasks for conventional BCI systems are generally imagination of body movements. The locations of EEG electrodes for conventional BCI systems are the motor sensory cortex areas or whole the brain area based on 10-20 standard EEG electrode system , .
In order to make efficient and economic BCI system, we need to minimize the number of EEG sensors. When we set the location of EEG sensors, location is closely related with the mental task. The illumination of the artifacts, however, is very difficult, because we don't know which the EEG signal is activated in accordance related with which specific mental task.
Our approach in solving this problem is using fMRI equipment. From fMRI, a higher level of spatial resolution data of brain activity area can be achieved than that of an EEG signal. The fMRI equipment measures the blood oxygenation level dependant (BOLD) signal , . The BOLD signal represents the level of oxygen consumed by brain activity, according to a mental task. Otherwise, the EEG sensor signal is data, which measure the electrical activity of neurons in the brain. Thus, the physical property of the data is different. There are, however, many reports, which state that the relation of BOLD and the local field potential (LFP) have an almost linear relationship -. The LFP is an electric signal, which is measured from the inside of the cortex and outside of the neuron cells. The physical property of the LFP signal is the same as that of the scalp-EEG signal. As a result, the fMRI experimental data may have a close relationship with the EEG sensor signals .
In general, the source of brain activity has been widely explained by three-dimensional dipole model . Thus, it seems to be reasonable to consider the dipole direction as well as the dipole source location. It is impossible, however, to estimate the dipole direction using fewer number of EEG sensors. In our case, the mental tasks to be considered are mainly related with surface brain activity, such as the secondary motor cortex area and the posterior parietal cortex area. Thus, we simply ignored the dipole directions of the source of brain activity. Based upon these observations, we have tried to find appropriate mental tasks, as well as to estimate the optimal location for the EEG sensors indirectly, by using fMRI equipment.
In Section 2, the fMRI experiment and data analyze will be described. In Section 3, the relation between fMRI data and EEG data and the EEG experiments, with analyze based upon these fMRI analysis results will be outlined. Section 4 provides a conclusion and further research.
II. fMRI Experiment and Data Analysis
A. fMRI experiment
The fMRI experiments were conducted with the cooperation of Brain Science Research Center (BSRC) at the Korea Advanced Institute of Science & Technology (KAIST). The experiments were conducted monthly in, 2005, and each experiment included two or three subjects. Among the eight subjects, one subject (subject A) participated in every experiment. The fMRI equipment included a 3T magnetic system. We set the time-to-repetition (TR) to 3000ms and the time-to-echo (TE) to 35ms. The motor imagery tasks were cued through the LCD project on the RF coil inside the gantry.
The fMRI experimental paradigm consisted of a condition state time for 12sec and a resting state time for 24sec. The repetitive mental tasks were repeated six times for each session. We analyzed the data by using Statistical Parametric Mapping (SPM) toolbox (FIL, London, England).
In order to find a suitable mental task, we executed various kinds of mental task experiment like imaging taste, and calculation of mathematical tasks, and imagination of good or bad experiences. We regarded suitable mental tasks as ones shows obvious brain activation with good recurrence and localization features in fMRI experiment data. After execution of the some kinds of mental task experiment and data analysis, we selected the imagination of body movement to focus metal task. Except for this mental task, we could not find similarities among the different subjects in the SPM analysis results.
We have applied mental task to map the two directions of a computer mouse point that can be controlled by the imagining of body movement, such as movement of the left finger and right fingers. Even though the mental task is simple, the exact meaning of the mental tasks was explained to the 8 subjects to avoid that subject misunderstand the mental task.
B. Analysis of fMRI experiment data
First, we check whether the observed data of the same mental tasks have similarities regarding different subjects and experimental periods. After observing the data, we could determine the brain activity in the supplementary motor area (SMA) for both left and right motor imagery. Figs. 1 and 2 show five months of data for subject A. Fig. 1 shows the SPM analysis results of the left-finger movement imagination experiment for the experimental periods between May and November. Fig. 2 shows the SPM analysis results the imagined of right-finger movement for each period. The first line of each figure indicates right-cerebral hemisphere activation and the second line shows left-cerebral hemisphere activation. The last line shows the dorsal view of the cortex activation. The SMA area is in the secondary motor cortex area, which plays an important role in planning body movements -. Furthermore, as shown in Figs. 1 and 2, the activation area near the SMA has clear localization and activation features. The data analysis results obtained from other subjects also show these SMA activation features clearly when mental task is imagination of body movement. By using these observations, we can conclude that the SMA area is always activated whenever subject performs imagery tasks.
[FIGURES 1-2 OMITTED]
Moreover, it is interesting to note that there are no prominent hotspots in the M1 area for both imaginations of right and left-finger movements. Moreover, it is difficult to find significant contralateral characteristics concerning the mental tasks in our experiment.
III. EEG Experiment and Data Analysis
A. EEG experiment
The EEG experiments were conducted using 16-channel BIO-PAC EEG acquisition equipment. The sampling rate of our experiment is 250Hz and the gain is 10,000. In the case of skin resistance, the amount was set below 5k ohm. There was little amount of 60Hz frequency power element in the Fourier transform data of the EEG data, and the amplitude of EEG is 30~60V, so the EEG data set was regarded as fair. In this experiment, we set the paradigms to similar format with the fMRI experiments to make similar conditions for both cases. The paradigm of the resting state time was set to 5 sec and the conditioning state time to 5 sec.
In this experiment, we set the experimental paradigm to a similar format for the fMRI experiments in order to standardize conditions. Our purpose for this EEG experiment was to verify the results of the fMRI experiment comparing those with the EEG sensor data. We choose the electrode positions: C3, C4, with Pz, Fz, AFz and Cz, of the 10-20 EEG electrode placement systems. The C3 and C4 areas are located in the primary motor cortex area (M1) especially, near the hand area in the penfield somatotopic map , and this area is usually referenced electrode location of BCI system whose mental task is imagination of finger movements. The SMA is the secondary motor cortex area (M2) and this area is located in middle of Cz and Fz, We executed EEG experiments with the imagination of left and right finger movements, resting. The experiment was repeated to ten times per each sequence.
B. Analysis of EEG experiment data
The two EEG features we regarded for offline analysis were the absolute values of 10~15Hz band filtering data and a variance of raw data. The 10~15Hz band EEG signal is related with * rhythm which is usually recorded from the motor cortex of the dominant hemisphere. It was possible to observe that variance of EEG signal decreased when the mental task was in a condition state. Fig. 3 shows the variance features. The interval of black line indicates the condition time interval. As shown in Fig. 3, it can be seen that the level of variance decreased when it is in condition interval, especially in channel 1 and channel 2, which is near the SMA area. We set the sampling interval time to 0.1sec and a 0% overlap time for extraction of the variance feature of the EEG data.
[FIGURE 3 OMITTED]
By using the absolute values of the filtered data and the variance of the raw EEG sensor signals, we implemented two LDA classifiers to discriminate the mental tasks. We regarded two kinds of discrimination method. One is distinguishing between resting state and condition state (imagination of left and right finger movement) and the other is distinguishing between imagination of left-finger movements and imagination of right-finger movements in the case of the condition state. Table 1 and 2 shows the result of analysis. Table 1 shows the results of discrimination between left finger motor imagery and right finger motor imagery. Table 2 shows the results of distinguishing between conditioned and resting state. Fig. 4 shows one example of LDA results. These are the result of distinguishing between imagined right and left-finger movement
[FIGURE 4 OMITTED]
From the results of the EEG signal, the motor-imagery mental task induces reliable distinct changes in the EEG signal features in the SMA area. The SMA area is located between the Fz and Cz. From the Tables 1 and 2, it is seen better discrimination performance in those areas. These results corresponded with those of the fMRI experiment analysis. Moreover, we can confirm that the location of the reference electrode is important. In the case of discrimination between left-finger motor imagery and right-finger motor imagery, the performance of the unipolar type EEG signal (earlobe-C3, earlobe-C4) is not as good as that of another case whose reference is in brain activity areas.
In this paper, we examined the reliable mental tasks and the location of EEG sensor to improve the performance of the BCI system using fMRI experiments and data analysis. We suggested the SMA as the suitable location for the BCI system, which was based on imagining finger movements. We mentioned that the proposed neurophysiological approach is highly necessary in order to improve the performance of the conventional BCI system, and also the fMRI experiments were able to provide opportunities to acquit reasonable answers to unsolved questions of the BCI system. We were able to estimate the location of reliable EEG sensors according to other proper mental tasks by using fMRI equipment, especially the non body movement imagination mental task, which is useful for the BCI system but proper EEG sensor locations of this are unknown.
This work was supported by the Korea Research Foundation Grant funded by the Korea Government (MOEHRD). (KRF-2005-202-D00459).
 Wolpaw, R., Niels Birbaumer, Dennis, J., McFarland, Gert Pfurtscheller, Theresa, M., Vanughan. "Brain-Computer Interfaces for Communication and Control", Clinical Neurophysiology, Vol. 112, pp. 767-791, 2002
 Garcia, G. Direct brain-computer communication through scalp recorded EEG signal. Lausanne EPFL Ph.D Thesis. 2004
 Niessing, J., Ralf A.W., Galuske, Boris Ebisch, Kerstin, E., Schmidt, Michael Niessing, Wolf Singer. "Hemodynamic Signal Correlate Tightly with Synchronized Gamma Oscillation", Science, 309 (5736). pp. 941-948, 2005
 Sameer, A., Sheth, Masahito Nemoto, Michael Guiou, Melissa Walker, Nader Pouratian, Arthur W., Toga. "Linear and Nonlinear Relationships between Neuronal Activity Oxygen Metabolism and Hemodynamic Response", Neuron, Vol. 42, pp. 347-355, 2004
 Nikos, K., Logothetis, Jon Pauls, Mark Augath, Torsten Trinath & Axel Oeltermann. "Neurophysiological Investigation of the Basis of the fMRI Signal", Nature, Vol. 412, 2001
 Christoph Mulert, Lorenz Jager, Robert Schmitt, Patrick Bussfeld, Olive Pogarell, Ulrech Hegerl. "Integration of fMRI and Simultaneous EEG: towards a Comprehensive Understanding of Localization and Time-course of Brain Activity in Target Detection", Neuroimage, Vol. 22, pp. 83-94, 2004
 Christoph M.Michel, Micah M,Murray, Goran Lantz, Sara Gonzalez, Laurent Spinelli, Rolando Grabe de Peralta. "EEG Source Imaging", Clinical Neurophysiology, Vol. 115, pp. 2195-2222, 2004
 Michael, S., Gazzaniga, Richard, B. ivry. George R. Mangun. Cognitive neuroscience, W.W. Norton .New York. London, New York, pp. 445-498, 2001
 Mark, F. Bear, Barry W. Connors, Micheal, A. Paradiso. Neuroscience exploring the brain, Lippincott Williams & Wilkins, Baltimore Maryland, pp. 43-44, 2000
 P. Dechent, K.-D., Merboldt, J. Frahm. "Is the Human Primary Motor Cortex Involved in Motor Imagery", Cognitive Brain Res, Vol. 19, pp. 138-144, 2004
Sang Han Choi birth in Korea in 1979. He is in master course in Kyungpook national university, electrical engineering and computer science. His major field is brain computer interface.
Minho Lee received the Ph.D. from Korea Advanced Institute of Science and Technology in 1995, and is currently an associate professor of School of Electrical Engineering and Computer Science, Kyungpook National University, Taegu, Korea, and visiting scholar in Dept. of Brain and Cognitive Science, Cambridge, MIT. His research interests include biologically inspired vision systems, brain computer interface and intelligent sensor systems. (Home page: http://abr.knu.ac.kr)
Sang Han Choi (1) and Minho Lee1, (2)
(1) School of Electrical Engineering and Computer Science, Kyungpook National University 1370 Sankyuk-Dong, Puk-Gu, Taegu 702-701, Korea email@example.com, firstname.lastname@example.org
(2) Department of Brain and Cognitive Science, Massachusetts Institute of Technology 77 Massachusetts Avenue, Cambridge, MA 02139, USA email@example.com
Table 1. The hit rate of single channel EEG data which distinguishes between an imagined left-finger motor imagery condition state and right-finger motor imagery condition state using LDA. Reference C3 C4 Pz 70% 90% Cz 80% 80% Fz 90% 90% AFz 90% 90% Ear lobe 60% 60% Table 2. The hit rate of single-channel EEG data which distinguishes between a conditional state and a state of rest using LDA Electrode location Left ear AFz Pz 79% 75% Cz 85% 75% Fz 81% 85% AFz 80%
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Brain computer interface|
|Author:||Choi, Sang Han; Lee, Minho|
|Publication:||International Journal of Computational Intelligence Research|
|Date:||Jan 1, 2007|
|Previous Article:||Development of the neuroinformatics base platform system: XooNIps.|
|Next Article:||Exchange options pricing with evolutionary neural-based fuzzy inference systems.|