ブレイン・マシン／コンピュータ・インターフェイス BMI/BCI
P1-1-230 リアルタイムMEGにおけるビームフォーミング技術と虚部コヒーレンス解析の実装 Implementation of beamforming technique and imaginary coherence analysis in real-time MEG ○大良宏樹1, 高野弘二1, 川瀬利弘1, 岩木直3,4神作憲司1,4 ○Hiroki Ora1, Kouji Takano1, Toshihiro Kawase1, Sunao Iwaki3,4, Lauri Parkkonen5, Kenji Kansaku1,4 国リハ研・脳機能部・脳神経1, 東工大・総理工・知シス2, 産総研・ヒューマンライフテクノロジー3, 千葉大・フロンティア医工学4, アールト大・低温・脳研究5 Sys Neurosci Sect, Dept of Rehab for Brain Func, Res Inst of NRCD, Tokorozawa, Japan1, DCISS, Tokyo Inst Technol2, Hum Tech Res Inst, AIST, Ikeda, Japan3, Res Ctr Front Med Eng, Chiba Univ, Chiba, Japan4, Brain Res Unit, Low Temp Lab, Aalto Univ, Espoo, Finland5 Real-time magnetoencephalography (rtMEG) is an emerging neurofeedback technology that could potentially benefit multiple areas of basic and clinical neuroscience. In the present study, we computed voxel-based imaginary coherence in real time using an rtMEG system in which we employed beamformers to improve localization of the signal source in the anatomical space. The acquired MEG signals were processed using signal-space projection (SSP) to reduce interference and the cortical sources were localized by applying a linearly-constrained minimum-variance (LCMV) beamformer. The putative functional connectivity between two ROIs was estimated by calculating the imaginary coherence between the source signals within them. We asked a healthy participant to attend to one of flickering visual stimuli (5 or 6Hz) in a MEG scanner (306-channel Elekta-Neuromag system; Elekta Oy, Helsinki, Finland). Feedback was presented visually and it was generated based on the obtained imaginary coherence between two ROI pairs: right posterior parietal cortex (PPC) and left visual cortex, and right PPC and right visual cortex. We showed that the participant successfully increased coherence between the PPC and the visual cortex during attending to the visual stimulus. The imaginary coherence between the right PPC and left visual cortex was greater than that between the right PPC and right visual cortex when the subject attended to the right flickering visual stimulus (p < 0.05). This finding suggests that our system is suitable for neurofeedback training and can be useful for practical brain-machine interface applications or neurofeedback rehabilitation.	P1-1-231 被験者間の荷重ベクトル上の事前分布による転移学習法 Using weight vector priors extracted from previous subjects to increase decoding performance ○ディブレクトマシュー1, 山岸典子1,2,3 ○Matthew de Brecht1, Noriko Yamagishi1,2,3 情報通信研究機構 脳情報通信融合研究センター1, ATR認知機構研究所2, JSTさきがけ3 CiNet, NICT, Kyoto, Japan1, ATR-CMC, Kyoto, Japan2, JST PRESTO, Saitama, Japan3 One major difficulty with brain-machine interfaces (BMI) is that a large amount of training data is required for calibrating the BMI every time it is used. Therefore, recently there has been growing interest in developing techniques to reuse data from previous training sessions. Here we will present results on using data from previous subjects to define a prior on the classifier weight vectors to help decrease the amount of training data needed. We found that using a prior from previous subjects can significantly increase decoding performance when little training data is available.In this study, we extended sparse logistic regression (SLR) with a prior that encodes the spatio-temporal dynamics of weight vectors extracted from subjects during previous decoding experiments. Twenty-four subjects participated in an experiment where a red and green grating was presented in either the left or right visual fields while brain activity was recorded with MEG. The decoding task was to determine which field the grating was displayed in based on the evoked response. For each subject, decoding accuracy using SLR was compared with accuracy of SLR extended with a prior constructed from the covariance information of weight vectors from the other twenty-three subjects. An average of 79.5 trials were available for each subject after artifact rejection. Ten-fold cross-validation was performed using 10% of the trials for training and the remaining trials for testing.We found that decoder accuracy increased by an average of 13.8 percentage points when prior information from other subjects was used (p<0.001, paired t-test). This shows that it is possible to reuse information from previous subjects to increase decoding performance, especially when there is only few data available for training. These results are important for the development of online BMI systems, where it is necessary to decrease the amount of time required for calibration.
P1-1-232 クラス内・クラス間分散に基づくBMI特徴量ベクトルの次元削減 Dimensionality reduction of BMI feature-vectors based on between-class and within-class variances ○恩田壮恭1, 椿田紘久1, 石山敦士1, 小野弓絵2 ○Masanori Onda1, Hirohisa Tsubakida1, Atsushi Ishiyama1, Yumie Ono2 早大・先進理工・電生1, 明治大・理工・電気電子生命2 School of Advanced Science and Engineering, Waseda Univ., Tokyo, Japan1, Dept. of Elec. and Bioinformatics, Sch. of Sci. and Tech., Meiji Univ., Kanagawa, Japan2 Dimensionality reduction of feature vectors originated from brain signals is essential to develop fast BMI system. Our purpose is to reduce the dimension of feature vectors from EEG data without declining the discrimination rate. Five healthy young-adult subjects participated in the study. We measured their EEG signals at a sampling rate of 128 Hz from 15 electrodes positioned at Fp1, Fp2, F3, Fz, F4, T3, C3, Cz, C4, T4, P3, Pz, P4, O1, and O2 according to the International 10-20 System. Subjects looked at the PC screen in front of them during the whole experiment. The PC screen showed fixation point (2s) at the center of screen followed by a blank screen (3s) for 10 times. Subjects were asked to image either task of (1) moving left hand, (2) moving right hand, or (3) doing nothing when the fixation point was present. Subjects repeated this trial 2 times for each task. We performed Fast Fourier Transform (FFT) analysis with the measured data and extracted power spectrum every 1 Hz between 6-20Hz for every 1s during the task period. An original, 15-dimensionnal feature vectors consist of the power spectrum at 15 electrodes. We determined the electrodes which showed the largest and the second largest difference among the three states. Specifically, we introduced the evaluation function to choose electrodes whose feature vectors have larger between-class variance and smaller within-class variance. Using a Gaussian kernel SVM, we discriminated the feature vectors which were reduced to 2 dimensions. We compared the discrimination rate between the cases in which (1) the 2-dimensional feature vectors were derived from the proposed method and (2) those were extracted by PCA, a typical dimensionality reduction algorithm in BMI. Our results demonstrated that the mean discrimination rate of the proposed method was 55.7±6.68% (S.D.), which was higher than that of PCA (51.3±5.21%). The result suggests that the proposed method may help developing fast and accurate BMI system.	P1-1-233 背側運動前野皮質が大きいほど脳波－ブレイン・コンピューターインターフェイスの成績が高い：Voxel-based morphometry解析を用いた相関解析 Dorsal premotor cortex structure correlated with control performance for an electroencephalography-based brain-computer interface: A voxel-based morphometry study ○笠原和美1,2, , 本田学1,3, 花川隆1,3,4 ○Kazumi Kasahara1,2, Charles S DaSalla1, Manabu Honda1,3, Takashi Hanakawa1,3,4 国立精神・神経医療研究センター　神経研究所　疾病研究第七部1, 首都大学東京　放射線科学域2, 国立精神・神経医療研究センター　脳病態統合イメージングセンター3, 科学技術振興機構　さきがけ4 Dept. of Functional Brain Research, Natl. Inst. of Neurosci, Tokyo1, Dept. of Radiological Sci., Tokyo Metropolitan Univ., Tokyo2, Integrative Brain Imaging Ctr., Natl. Ctr. of Neurol. and Psychiatry, Tokyo3, PRESTO, Japan Sci. and Technol. Agency, Kawaguchi4 The potential of brain-computer interfaces (BCIs) to replace lost neuronal function in the form of neuroprostheses has been widely studied. Moreover, BCIs have gained recent attention as a possible means to induce beneficial neuronal plastic changes via neurofeedback training. However, although BCI performance considerably varies among subjects, the factors affecting BCI performance are poorly understood. Therefore, we investigated the relationship between performance of an electroencephalographic (EEG) mu rhythm-based BCI (EEG-BCI) and brain structure. Twenty-four healthy subjects were instructed to control a computer cursor using left- and right-hand motor imagery via the EEG-BCI. EEG data were recorded using an 11-channel head cap and amplifier. Left- and right-hemispheric mu band powers were then calculated from the EEG data stream and converted into a control signal for cursor movement. EEG data showed that subjects were able to modulate their mu band powers and control the BCI with accuracies significantly above the chance level. Following this experiment, subjects underwent T1-weighted three-dimensional structural imaging, using a 3 Tesla MRI scanner. The MRI data were subjected to voxel-based morphometric analysis, using BCI control performance as an independent variable. We identified a correlation between EEG-BCI performance and left dorsal premotor cortex (PMd) volume. This finding indicates that control performance for the present EEG-BCI was higher in subjects with greater left PMd volume. The PMd is known to be involved in motor imagery, and we speculate that subjects with greater left PMd volume were more adept at imagining hand movement. Furthermore, we presume that it was necessary for the subjects to switch quickly between left- and right-hand motor imagery to achieve maximal performance. Therefore, the PMd may be the hub where switching between left- and right-hand motor imagery occurs.[Kasahara and DaSalla contributed equally to this work.]
P1-1-234 SSVEP-BCIのためのLEDコントラストと刺激時間長の検討 Determination of appropriate duration and contrast of flicker stimuli for SSVEP-based BCI ○佐伯謙太朗1, 浦野貴文1, 池本健亮1, 小野弓絵1 ○Kentaro Saeki1, Takahumi Urano1, Kensuke Ikemoto1, Yumie Ono1 明治大院・理工・電気1 Grad Sch of Sci&Tech Meiji Univ,Kanagawa1 Steady-state visual evoked potential (SSVEP) is one of the easy and promising techniques of brain-computer interface which require little training, although an appropriate characteristics of flicker stimulation to evoke strong SSVEP has not yet well determined. Seven subjects without perceived vision problem (20-40 years old, mean age 24.1) participated in the experiment. We recorded electroencephalogram from the occipital area Oz according to the international 10-20 system. Subjects watched a while-LED light flickering at either 6 or 8Hz for 70s which was built on a breadboard. We recorded SSVEP in different contrast conditions, in which the surface of the white-color breadboard was covered with black cardboard (high-contrast) or not (low-contrast). The data were split into several segments with different durations of 5, 6, 8, and 10s. With each segmented data we calculated the spectrum power values at 6 and 8 Hz, which constitute the training set for linear discriminant analysis (LDA) classifier. Cross-validation indicated that the best discrimination rate was achieved when the training set was derived from the data segmented by the longest duration (10s, 83.3%). However, the discrimination rate was significantly larger than the chance level even when the training set was calculated from the data segmented by the shortest duration (5s, 78.0%). We did not find any statistically significant improvement in the discrimination ratio regarding to contrasts. The mean discrimination rates were 74.1% and 78.2% with high-contrast and low-contrast conditions, respectively. These results suggest that the light intensity was strong enough to evoke robust SSVEP signals regardless of the background color. Further research will be needed to investigate the discrimination rate when the training set was calculated from the data segmented by shorter duration than those used in the current study.	P1-1-235 低周波数の点滅光は強くSSVEPを誘発する Increased SSVEP amplitude with flicker stimulus of lower frequency ○浦野貴文1, 佐伯謙太朗1, 池本健亮1, 小野弓絵1 ○Takafumi Urano1, Kentarou Saeki1, Kensuke Ikemoto1, Yumie Ono1 明治大院・理工・電気1 Grad Sch of Sci&Tech, Meiji Univ, Kanagawa1 Steady-state visual evoked potential (SSVEP) is one of the easy and promising techniques of brain-computer interface (BCI). SSVEP-BCI has an advantage of little training required, however an appropriate frequency band of flicker stimulation to evoke strong SSVEP has not yet well determined. Using LED flicker stimulator, we investigated the difference in power spectrums of the visual-evoked potential when subjects watched white LED light flickering at 6, 8, 13, 15, 23, and 27 Hz. Five subjects without perceived vision problem (21-34 years old, mean age 24.3) participated in the experiment. We recorded electroencephalogram (EEG) from the occipital area (O1, O2, and Oz according to the international 10-20 system). Subjects watched each flicker light for 30s. We determined the amplitude of SSVEP signal as the maximum peak value of the FFT power spectrum at Oz between the frequency bands ±0.2Hz around the exact frequency of visual stimulation. SSVEP signals obviously appeared with the low-frequency flickers in most of the subjects. However, the amplitude of SSVEP signals decreased as the stimulation frequency increased, and all subjects failed to show SSVEP spectrum peak above the noise level at 23 and 27 Hz stimuli. The mean and the standard deviation of the amplitudes of SSVEP signals were 6523±4048, 7091±5673 , 5683±2185, 5214±3109, 3037±825, and 3636±813 uV2 for 6, 8, 13, 15, 23, and 27 Hz flickers, respectively. These results suggest that the low frequency flicker (6,8Hz) contributes to better controllability of SSVEP-based BCI.
P1-1-236 指タッピング課題におけるNIRS酸素化ヘモグロビン濃度変化信号は脳波ERSと関連する Oxy-hemoglobin activity correlates with gamma-band ERS: simultaneous NIRS and EEG measurement during finger tapping task ○辻本翔1, 李基準2, 小野弓絵1 ○Sho Tsujimoto1, Kijoon Lee2, Yumie Ono1 明治大院・理工・電気1, 南洋理工大学2 Grad Sch of Sci and Tech, Meiji Univ, Kanagawa1, Div. of Bioengineering, Nanyang Tech. Univ., Singapore2 Although recent research suggests that near-infrared spectroscopy (NIRS) and electroencephalography (EEG) complement each other to improve the performance of noninvasive Brain Computer Interfaces (BCI), the relationship between signals from these two modalities needs to be elucidated to further utilize the different characteristics of these signals in the application of BCI. We therefore simultaneously recorded NIRS and EEG and investigated how metabolic changes of oxy-hemoglobin (Oxy-Hb) concentration correlate with the electrophysiological sensorimotor rhythms during actual motor execution. Five subjects who are right-handed performed two sessions of finger tapping task using either left or right hand, in a block design with 5 s of finger tapping followed by 15 s of rest repeated 10 times. We bilaterally placed 4x4 optical probes for NIRS and 5 EEG electrodes on the scalp around the C3/C4 in the international 10-20 system. We found that correlation between NIRS and EEG signals do differ depending on the subject. One subject shows that 10 NIRS signals and that of EEG while this subject does finger tapping ten times are well correlated each other with the Pearson's coefficient of correlation which are -0.688 in left hemisphere and -0.782 in right hemisphere as a result of correlation analysis by SS?. Other subjects do not show correlation between them in either side of finger tapping. Considering that we can reduce time and effort to advance BCI research because we do not need to apply correlated factors to BCI if we simultaneously recorded NIRS and EEG before input into BCI and find that there is subject whose signals of NIRS and EEG are correlated each other during task.	P1-1-237 新奇モダリティ獲得の可能性 Capability of rats to utilize a new modality ○乘本裕明1, 松木則夫1, 池谷裕二1 ○Hiroaki Norimoto1, Norio Matsuki1, Yuji Ikegaya1 東京大学大学院　薬学系研究科　薬品作用学教室1 Dept Pharmaceutical sci, Univ of Tokyo, Tokyo1 In human and other mammals, the sensory organs play a critical role in the acquisition and processing of information. Although our current senses may be sufficient for living, there are numbers of other natural signals we cannot sense, such as geomagnetism, ultrasonic waves, and radiation. Here, we developed a microsensor that detects the magnetic field and applies electrical stimulation to the brain, depending on the magnetic field. We implanted this sensor in either primary visual cortex or barrel cortex. The rats were able to solve spatial navigation tasks dependent on the geomagnetic field, only when the sensor was turned on. These data indicate the possibility that the direct signals from the artificial sensor are integrated into higher brain function, such as memory and learning, along with the pre-existing modalities. Thus this work opens a new avenue that allows the brain to interact the external world and expands the potential adaption of the brain.
P1-1-240 一次運動野の皮質脳波を用いた腕軌道のデコーディング Decoding of arm movement from electrocorticogram in primary motor cortex ○陳超1, 辛徳2, 中西康彦2, 神原裕行2, 吉村奈津江2, 渡辺秀典3, 南部篤4, 伊佐正3, 西村幸男3, 小池康晴1,2 ○Chao Chen1, Shin Duk2, Yasuhiko Nakanishi2, Hiroyuki Kambara2, Natsue Yoshimura2, Hidenori Watanabe3, Atsushi Nambu4, Tadashi Isa3, Yukio Nishimura3, Yasuharu Koike1,2 東京工業大学　総合理工学研究科　物理情報システム　専攻1, 東京工業大学　精密工学研究所2, 生理学研究所　認知行動発達機構研究部門3, 生理学研究所　生体システム研究部門4 Department of Information Processing, Tokyo Institute of Technology, Yokohama, Japan1, Precision and Intelligent Laboratory, Tokyo institute of Technology, Yokohama, Japan2, Department of Developmental Physiology, National Institute for Physiological Sciences, Okazaki3, Department of System Integrative Physiology, National Institute for Physiological Sciences, Okazaki4 As prospective neural signals used in Brain machine interface, electrocorticogram (ECoG) signals has been in focus during recently years. A number of works had been done, such as classification of direction of arm movement, natural grasp types, regression of two or three dimensions arm trajectory, and muscle activities in time series. Despite such successes, however, in the view of engineering, there still remains considerable work to be done before a high performance neural prosthetic can be realized. In this paper, we proposed an algorithm to decode hand trajectory from ECoG signals recorded from primary motor area in primates. In the phase of feature selection, ECoG signals were band-pass filtered into separate sensorimotor rhythm bands, z-score normalized, and smoothed with a Gaussian filter. Then, Partial Least Squares Regression Method was used to train a predicting model to decode hand trajectory. Finally, performance of predicting model was test by validation dataset. The result showed hand trajectory has been decoded successfully with high performance by this new method. Furthermore, we investigated how many electrodes were necessary in predicting hand trajectory and compared the performances of predicting model according with the numbers of electrodes used. The results showed that hand trajectory can be decoded by using three ECoG signals. Finally, we discussed the optimal strategy in selecting electrodes.	P1-1-241 BCIへの応用を目指した運動想像トレーニングによる脳の可塑性 Change of brain activity effect from motor imagery training: application to EEG-based brain machine interface ○小川剛史1, 東良樹2, 石井信1,2 ○Takeshi Ogawa1, Yoshiki Azuma2, Shin Ishii1,2 国際電気通信基礎技術研究所　動的脳イメージング研究室1, 京都大学工学部電気電子工学科2 Dept. Dynamic Brain Imaging, ATR, Kyoto1, Faculty of Engineering, Kyoto Univ.2 Recent studies show that electroencephalography (EEG), one of non-invasive brain measurement modalities, has great potential for brain computer interface (BCI). Although a number of EEG-based BCI studies focus on decoding of mental states during motor imagery (MI) tasks, it is difficult for subjects to generate stable EEG signals associated with the MI without repeated MI training. In this study, we investigated the effects of daily MI training in terms of brain activity's change. Four right-handed healthy volunteers were participated to make it possible to imagine movements of their right and left hands in response to visual cues for four consecutive days. Through the daily MI training, the power in μ frequency bands (8Hz-13Hz) decreased relative to the baseline level (event-related desynchronization: ERD) on the contralateral motor area (C3 channel: left motor area; C4 channel: right motor area). But, there were no consistent inter-day changes in that area. ERD on Cz channel was also found during both of hand movement imagery trials (right/left hand) and this ERD was decreased along the four training days. In particular, the ERD with a broad band (10-25Hz) appeared in earlier days, while the ERD with a narrow band (10-15Hz) on the last day. Such changes in the frequency domain suggest that the subject's skill could be consolidated in the middle of motor areas such to be associated with the MI through the MI training. In contrast, the classification accuracy was not changed significantly across the training days, which is consistent with previous reports. A possibly reason is that the classifier as usual assumed stationarity of the training data, i.e., it did not consider the changes in neuronal representation of the MI due to the MI training. In addition, we will compare fMRI signal changes and the EEG signal changes above, both due to the MI training.
P1-1-242 脳磁図を用いた日常動作における手先の動きの再構成 Decoding hand movements in everyday activities from magnetoencephalography ○平山淳一郎1, 鹿内友美1,2, 中村泰1,3, 前田新一1,2, 竹之内高志1,4, 兼村厚範1, 川鍋一晃1, 石井信1,2 ○Jun-ichiro Hirayama1, Yumi Shikauchi1,2, Yutaka Nakamura1,3, Shin-ichi Maeda1,2, Takashi Takenouchi1,4, Atsunori Kanemura1, Motoaki Kawanabe1, Shin Ishii1,2 ATR・脳情報解析1, 京都大院・情報2, 大阪大院・基礎工3, 公立はこだて未来大・システム情報4 ATR Neural Information Analysis Labs, Kyoto1, Grad School of Informatics, Kyoto Univ, Kyoto2, Grad School of Eng Sci, Osaka Univ, Osaka3, School of Systems Information Sci, Future Univ Hakodate, Hakodate4 Decoding hand movements from non-invasive brain signals is a fundamental research issue in brain machine interface (BMI). Recent studies have shown that it is possible to reconstruct precise single-finger movements from magnetoencephalography (MEG). The present study aims at generalizing this result to more coordinated behaviors of hand and fingers like what we conduct in everyday life. For this purpose, we conducted an MEG experiment in which one healthy right-handed subject was asked to perform one of ten different kinds of right-hand actions each associated with a daily object like a mug or a stapler. The movement was executed within two seconds in each trial, and total 810 trials were collected in nine sessions of continuous MEG recording. The raw MEG signals were band-pass filtered at 0.5-30Hz, and typical artifacts were removed by ICA. A motion capture device also measured the positions of eight optical markers attached to the right hand, of which three markers were only used to compensate rotation of wrist. For decoding, 100msec time windows of marker positions were sampled by sliding windows of 10msec overlaps, and each of them was predicted from 100msec time windows of MEG preceding 50msec to the target time window. The probabilistic canonical correlation analysis (CCA) was used for associating the two time windows, and the decoding performance was evaluated using a cross-validation procedure. As a result, the highest cross-validated canonical correlation was about 0.6 in some objects, and the markers achieved high correlation coefficient over 0.7 between true and predicted positions, except for those on thumb which consistently exhibited relatively small correlations. This result showed a possibility that realistic coordinated actions of hand and fingers can be decoded from non-invasive brain signals, which would improve the low throughput of current non-invasive BMI systems.	P1-1-243 海馬シータ波による人工環境制御 Artificial environment controlled by the hippocampal theta in rats ○高野裕治1,2, 高橋伸彰2,3, 請園正敏2,4, 廣中直行5 ○Yuji Takano1,2, Nobuaki Takahashi2,3, Masatoshi Ukezono2,4, Naoyuki Hironaka5 NTTコミュニケーション科学基礎研究所1, CREST, 科学技術振興機構2, 関西学院大学3, 明治学院大学4, 三菱化学メディエンス5 NTT Communication Science Laboratories, Kanagawa1, CREST, Japan Science and Technology Agency, Kanagawa2, Kwansei Gakuin University, Hyogo3, Meijigakuin University, Tokyo4, Mitsubishi Chemical Medience, Kumamoto5 The hippocampal theta is a sinusodial-like wave at about 4-10 Hz in the local field potential recorded in the hippocampus. Because the theta in rats appears during exploring or rapid eye movement sleep, the functional role has been thought as arousal state in the brain. Many previous studies in rats showed that it was possible to discriminate between waking-state and sleeping-state by using the appearance rate of hippocampal theta and body-movements. For example, an increment of theta-rate means dark cycle for rats in most cases, and a decrement of theta-rate means light cycle. Then, we developed an artificial environment controlled by the hippocampal theta in rats. First, we recorded the hippocampal local field potential during 24 hours in freely moving rats, and then analyzed the appearance rate of the theta. Next, we turned the light off after the increment of the theta-rate, and turned the light on after the decrement of the theta-rate. Using some levels of the increment and decrement of the theta-rate, most suitable light and dark cycle has been examined. The animal behaviors in the environment controlled by the theta has been compared with just 12 hours light and dark cycle. We will discuss the application potentiality of brain-artificial environment interface in the future.
P1-1-244 飽きを感じた時の脳波コヒーレンス変化 The change in EEG coherence when you are bored ○夏目季代久1, 片山冬馬1 ○Kiyohisa Natsume1, Touma Katayama1 九州工業大学院生命体工学研究科1 Dept. of Brain Sci. and Eng., Grad. Sch. of Life Sci. and Sys. Eng., Kyusyu Inst. of Tech., Kitakyusyu, Japan1 English rhythm instruction materials (RIM) encourage one to study English rhythm. There are seven lessons in the RIM. In a lesson, you have to speak loud following the English teacher's song looking at the phrases of the song in the text. You have a RIM lesson once a day for five days. During the learning the power of theta band (4-8 Hz) is increased at the frontal region [1]. When you repeat a lesson several times in a day, you become bored. It was reported that the power of alpha (8-14 Hz), beta (14-30 Hz) and gamma (30-50 Hz) bands started to decrease at all brain regions before the subjects felt the bored feeling. On the other hand, theta power did not change [2]. In this study, we studied the relationship between electroencephalogram (EEG) coherence and the bored feeling. Six healthy male subjects (average years 23.4 ± 1.67; mean ± S. D.) had to repeat the same RIM 15 times, and to raise their hands when they felt bored. It took about 18 to 19 min. Subjects repeated RIMs recording EEG without a break. EEG at eight locations (Fz, Cz, Pz, Oz, T3, T4, F3 and F4) in an international 10/20 system was measured. The coherence between the two locations which were selected from eight recording sites was calculated using Matlab (Mathworks Inc., USA) software. It was calculated from EEG signal for 100 sec before and after a subject raised his hand. The coherences were calculated at theta, alpha, beta, and gamma bands. In results, for five subjects, the coherence at all bands decreased on the electrode pairs of head's midline (Fz, Cz, Pz and Oz). These results suggest that the decrease in the coherence of all frequency bands on head's midline can also be the sign for the subject's bored feeling. [References] [1] Nakano H., Yoshida, N. and Natsume, K. (2007) Language Education & Technology, 44:155-167. [2] Katayama T. and Natsume K. (2012) Journal of signal processing, Vol. 16, 6:637-641.	P1-1-245 ニューロコミュニケーター 2.0 － カスタムメードのヘッドギアを用いた実用的BMIシステム Neurocommunicator 2.0 as a practical BMI system with the custom-made headgear ○中村美子1, 工藤泰彦2, 長谷川良平1 ○Yoshiko Nakamura1, Yasuhiko Kudou2, Ryohei P. Hasegawa1 (独)産業技術総合研究所 ヒューマンライフテクノロジー研究部門1, (独)産業技術総合研究所 イノベーション推進本部2 National Institute of Advanced Industrial Science and Technology,Human Technology Research Institute1, National Institute of Advanced Industrial Science and Technology,Research and Innovation Promotion Headqarters2 There is recent world-wide interest in developing brain-machine interface (BMI) that enables direct connection between the brain and a machine. It is expected that the BMI works as advanced assistive technologies. We have been developing the "neurocommunicator", a cognitive BMI system for people with severe motor disabilities who cannot speak and write. We focused on a P300 potential as a useful neural signal when choosing one of pictograms that contain messages. Close-fitting cloth caps are often used to detect such a signal, which is strongest over the parietal cortex. We also used to use those caps for the EEG recording in the previous version of the neurocommunicator (version 1.0). They, however, tend to squeeze the user's head, and were not suitable for long-term recording. In this study, we introduced the custom-made plastic headgear in order to avoid such a squeezing. We examined the performance of the headgear with 9 normal subjects. The new system (neurocommunicator 2.0) actually improved both usability and accuracy, suggesting it as a practical solution for communication aids.
P1-1-246 NIRSを事前情報としEEGから推定された皮質電流からの空間注意のデコーディング Decoding of spatial attention from cortical currents estimated from EEG with NIRS prior ○森岡博史1,2, 兼村厚範1, 森本智志1, 吉岡琢1, 川鍋一晃1, 石井信1,2 ○Hiroshi Morioka1,2, Atsunori Kanemura1, Satoshi Morimoto1, Taku Yoshioka1, Motoaki Kawanabe1, Shin Ishii1,2 ATR脳情報解析研究所1, 京都大学大学院情報学研究科2 Neural Information Analysis Laboratories, ATR, Kyoto, Japan1, Graduate Shool of Informatics, Univ of Kyoto, Kyoto, Japan2 Electroencephalography (EEG) has been widely used for Brain Machine Interface (BMI) in real environments because of its high portability. However, EEG signals suffer from volume conduction effects and are usually mixtures of neural activities from broad area, some of which may not be related to the task targeted by the BMI, and then cause the accuracy of BMI to decrease. In this study, we aimed to solve this problem by estimating cortical source currents from EEG sensor signals by incorporating the information of brain activity regions measured by near-infrared spectroscopy (NIRS) using Variational Bayesian Multimodal EncephaloGraphy (VBMEG); after the source estimation, the BMI decoder was estimated by sparse logistic regression (SLR). VBMEG allowed us to isolate current sources on the basis of their cortical locations. In addition, by using SLR, we can select relatively small number of current sources relevant to the task of our target. In particular, we focused on the decoding of spatial attention. Covert spatial attention is a standard paradigm for BMI because it would be convenient for subjects to control objects in the visual domain such as a computer cursor. In the experiments, eight healthy subjects performed a covert spatial attention task (Left or Right), and the subjects' brain activities were simultaneously recorded with EEG and NIRS. The results showed that, from estimated current sources, subjects' attended directions were well decoded with the accuracy significantly higher than that from solely EEG sensor signals. Moreover, spatial distribution of weight values of the classifiers revealed that classifiers obtained by SLR placed informative current sources around the intraparietal sulcus (IPS), which is known to be involved in spatial attention. These results suggest that our method improves the accuracy of BMI by isolating cortical currents from EEG sensor signals and by selecting informative current sources for decoding in an effective manner.	P1-1-247 視覚野刺激型人工視覚のための多点電流刺激VLSIの開発 A multi-channel current stimulator fabricated with Very Large Scale Integrated Circuit technology for cortical prostheses ○亀田成司1, 林田祐樹2, 田中宏喜1, 秋田大2, 八木哲也2 ○Seiji Kameda1, Yuki Hayashida2, Hiroki Tanaka1, Dai Akita2, Tetsuya Yagi2 大阪大・MEIセンター1, 大阪大院・工2 MEI Center, Univ of Osaka, Osaka1, Grad Eng, Univ of Osaka, Osaka2 In visual prostheses by means of cortical stimulation, integrated electronic devices that can generate spatio-temporal patterns of electrical stimuli with a large number of electrodes are desirable. We fabricated a prototype of a multi-channel current stimulator chip with 0.25-μm high-voltage CMOS VLSI technology. The chip has 20 output channels which are driven by five current generators arranged in parallel, each of which control four output channels independently in time-sharing mode. Stimulus parameters of a current pulse, e.g., amplitude and duration, can be controlled separately for each output channel by digital codes stored in built-in registers. The combination of anode and cathode electrodes for passing stimulus current can be changed online. The chip was able to generate current pulse amplitude up to approximately ±100 μA/phase in high-impedance glass microelectrode (~1 MΩ), and the waveform of pulse was distorted to some extent. By using the voltage-sensitive dye imaging technique, we measured neural responses to extracellular current stimuli generated by the stimulator chip in the rat visual cortex V1 in vivo. When a train of five or ten consecutive biphasic current pulses (9 μA/phase amplitude, 200 μs/phase duration, 5 ms inter-pulse interval) was applied via the microelectrode, fluorescent signal corresponding depolarizing responses were observed around the stimulation site within 5-10 ms after the stimulation onset. The signal spread to peripheral regions in the visual cortex V1. The response amplitude peaked at 20 ms after the onset and maintained around the peak level for the stimulation period, then declined below the resting level. The response signal was similar to those typically induced with a commercially available stimulus isolator. The chip fabricated in the present study is useful for cortical prosthetics because of its compact and low power architecture, and extendable channel counts.
P1-1-248 サル到達把持運動における皮質表面電位と皮質内局所電位の上肢運動・筋電位の推定精度の比較 Comparison of decoding accuracy of arm movements and electromyogram of arm muscles between electrocorticogram and local filed potentials during reach-to-grasp movements in a monkey ○渡辺秀典1,2, 佐藤雅昭2, 南部篤3,4, 西村幸男1,4,5, 川人光男6, 伊佐正1,4 ○Hidenori Watanabe1,2, Masa-aki Sato2, Atsushi Nambu3,4, Yukio Nishimura1,4,5, Mitsuo Kawato6, Tadashi Isa1,4 生理学研究所・発達生理1, ATR 脳情報解析研究所2, 生理学研究所 統合生理3, 総研大4, さきがけ5, ATR 脳情報研究所6 Dept Dev Physiol, NIPS, Okazaki1, ATR Neural Infor Analy Lab, Kyoto2, Dept Inte Physiol, NIPS, Okazaki3, SOKENDAI, Hayama4, PREST, Tokyo5, ATR Compute Neurosci Lab, Kyoto6 Both intracortical local field potentials (LFPs) and electrocorticogram (ECoG) have been shown to carry reliable information about arm movements and electromyogram (EMG). In general, LFPs are supposed to reflect neuronal activities in the deeper layers more substantially than ECoG. However, in our previous study, we found that considerable amount of signal in the deeper layers was also included in ECoG (Watanabe et al., 2012). It is still unclear whether ECoG carries less information about arm movements than LFPs. Here, we directly compared the decoding accuracy of arm movement kinematics and arm EMGs estimated by simultaneously recoded LFPs and ECoG in the monkey primary motor cortex (M1). ECoG was recorded from 32 positions in the surface around the hand and arm regions of the M1 during the reach-and-grasp task. LFPs were recorded from 2 positions in the superficial and deep layers of the same regions using four 2-channel electrodes. Kinematics of arm movements was continuously recorded using 10 different parameters. EMGs were recorded from 10 muscles in the hand and arm. Kinematics and EMGs were reconstructed from the powers at 3 frequency-bands (1-5, 10-35 and 80-170 Hz) in the single channels of the ECoG and LFP using a sparse linear regression. When the recording sites of LFPs and ECoG were close, the decoding accuracy was comparable (kinematic, R=0.79-0.43; EMG, R=0.83-0.45). Best decoding accuracy from an ECoG channel among the ECoG array were significantly higher than the best channel from the LFP channels for the reconstructions of kinematics of 7 out of the 10 different parameters and 6 out of 12 EMG recordings. ECoG channels with best decoding accuracy were mostly located in the M1 and primary somatosensory cortex. These results suggest that ECoG carries rich information comparable to LFPs and can decode arm movements and EMGs with high accuracy.	P1-1-249 音楽嗜好課題における脳波 The electroencephalogram in preference judgment task of music ○小柳諒輔1, 小島昇1, 夏目季代久1 ○Ryousuke Koyanagi1, Noboru Kojima1, Kiyohisa Natsume1 九州工業大学大学院 生命体工学研究科 脳情報専攻1 Dept. of Brain Sci. and Eng., Grad. Sch. of Life Sci. and Sys. Eng., Kyusyu Inst. of Tech., Kita-kyusyu, Japan1 The daily behavior of the person is affected the subjective preference. Recently the study on the electroencephalogram (EEG) which is related to the preference has been started. Most of the studies so far have been focus on the visual preference. The studies clarify that prefrontal theta and occipital gamma bands are involved in the preference. On the other hand, music has a relaxation effect, and can reduce our stress (Naito, 2006). Recently musical therapy is used for medical care and the welfare (Ishihara, 2008). The effect is maximized when they listen to the favorite music. Thus to know the individual's music preference is also important. In the present study, the music preference was studied using electroencephalogram (EEG). Five healthy males (23.5±0.5 years) participated in the experiment. They listened to the music EEG for 30-35 sec, thought for 10 sec, and then decided the preference in four levels. EEG was recorded throughout the experiment. It was analyzed using fast Fourier transform. The mean power levels of theta (4-8 Hz), alpha (8-12 Hz), beta (12-30 Hz), and gamma (30-50 Hz) were calculated while the subject listened to the music, and while they thought the choice. In addition, cross-validation of the power levels in one dimension was done using the linear discriminant analysis (LDA).In the preference thinking period, the averaged power of the prefrontal theta band at "like" choice was higher than at "dislike" choice. The difference was not significant. On the other hand, both during listening to music and preference thinking period, the averaged power of the occipital gamma band tended to be higher at "dislike" choice than at "like" choice. The difference was not significant. The cross-validation result showed that the three highest correct answer rates were obtained using the occipital gamma, the prefrontal alpha, and the occipital gamma bands. Hence, the occipital gamma bands may participate in the "like" and "dislike" emotion.
P1-1-250 色知覚と定常性視覚誘発電位の振幅の関係の研究 A study of the relationship between color perception and the amplitude of steady-state visual evoked potential ○廣瀬秀顕1, 宮城大輔1, 小池康晴2 ○Hideaki Hirose1, Daisuke Miyagi1, Yasuharu Koike2 株式会社アイシン・コスモス研究所1, 東京工業大学ソリューション研究機構2 R&D Dept., Aisin Cosmos R&D Co., Ltd., Kariya, Japan1, Solution Science Research Laboratory, Tokyo Institute of Technology, Yokohama, Japan2 Previous studies have shown that the amplitude of steady-state visual evoked potential (SSVEP) is affected by the color of visual stimulus. Then, it was assumed that color perception is a crucial parameter for eliciting SSVEP. By the way, about 5-8% of the population has some sort of inherent color blindness. Now, there are about 3 million patients with the color blindness in Japan. Furthermore, aging and eye diseases can cause acquired color blindness. To diagnose these color blindness, Pseudoisochromatic plates and anomaloscope has been used. We thought that, if SSVEP includes color perception information, we might diagnose the color perception level of patient directly. The aim of this study was to elucidate the relationship between color perception and the amplitude of the steady-state visual evoked potential. In experiment, subjects with normal color vision were required to gaze visual stimuli (a red, green or blue-colored rectangle) for five seconds. The visual stimuli were flickered at 20 or 60 Hz. The background color of the visual stimuli was black. And the subjects were required to gaze the visual stimuli with the naked eye (condition 1) or with a glass which has a band-pass color filter (red, green or blue; condition 2, 3 or 4) for making color blindness artificially. Electroencephalogram (EEG) was measured on the scalp over the visual cortex. Average amplitude of SSVEP was calculated from the measured EEG. From the experiment, compared with the condition 1, it was observed that the amplitude of SSVEP increased for the red-colored visual stimulus and decreased for the green and blue in the condition 2. Similar result was obtained even if the flicker frequency was different. The same tendencies were also obtained in the condition 3 and 4. From the result, it was suggested that SSVEP is useful as a diagnostic tool to reveal the color perception ability of person. By using SSVEP, more accurate color perception diagnosis may be realized.	P1-1-251 EEG-based control of a Wearable Robot: How passive movements induced by the robot influence the brain activity and the control performance ○Giuseppe Lisi1,2, Tomoyuki Noda1, Masa-aki Sato3, Jun Morimoto1 ATR Computational Neuroscience Laboratories1, Nara Institute of Science and Technology2 A wearable exoskeleton robot is a mechatronic system designed around the shape and function of the human body, which can be employed, among others, in the assistance of disabled persons, rehabilitation and neuromotor control research. Our aim is to design a non invasive EEG-based control system for a wearable robot, which must be reliable and robust against the electrical and mechanical noise coming from the robot and against possible interfering brain activities generated by the passive movements induced by the exoskeleton. A voluntary or passive movement or a motor imagery task very often result in a change of the ongoing EEG in form of an event-related desynchronization (ERD) or synchronization (ERS). Given their repeatability, these phenomena can be reliably used as input to the wearable robot. In our study, Time-Frequency analysis and Sparse Logistic Regression are respectively employed for the online extraction and classification of the features related to the ERD/ERS. In order to investigate how well the classifier performs without the influence of the exoskeleton, in the first set of experiments the subjects were not attached to the robot and performed different types of mental tasks to control it from distance. In the second set of experiments one of the subjects' leg was attached below the knee to a 1DoF robot and the subjects performed the same tasks as in the previous experiment in order to provide an input to the classifier. In this case the robot was not controlled by the subject, but it continuously and automatically moved at low frequencies, simulating a part of a walking pattern and inducing a passive movement throughout the whole experiment. In this way we were able to investigate the robustness of our classifier when the brain activity is affected by the passive movements induced by the robot. Moreover, in order to analyze how electrical and mechanical noise could interfere with the EEG signal we tested different types of actuation of the robot.
P1-1-252 Event related dynamics of EEG oscillations in cued motor imagery ○Archana Singh1, T. Ogawa1, T. Yoshioka2, J. Hirayama1, M. Maruyama1, S. Ishii1 ATR/NIA/DBI1, ATR/NIA/CBI2 Using mental simulation of the movements we may produce sensation in the brain which would also be evoked during execution of the movements. Therefore, motor imagery has become a standard control tactic in brain machine interface (BMI). Most decoding methods assume that motor imagery evokes specific EEG oscillatory patterns (e.g., mu and beta rhythms) in certain (e.g., contralateral or bilateral) motor areas. However, very few studies have investigated such assumptions thoroughly; the question whether imagery (and motor execution) motor involves contralateral and/or bilateral areas still lacks a clear consensus. In addition, very few studies localized the decoded EEG patterns to their cortical source. In this study, we investigate EEG signals from a BMI training experiment to provide better neuropsychological understanding of the signal and develop better BMI decoding methods. Subjects performed motor imagery by sequentially tapping the fingers of their left and right hands (randomized over trials). We computed ERD/S using wavelet-decomposition, which is robust to non-stationary EEG signals and offers much finer details of EEG dynamics varying in time and frequency domains. The cortical sources of the detected oscillations are determined by Variational Bayesian Multimodal Encephalography method incorporating subject's fMRI. Our analysis reveals a readiness potential with significant contralaterally dominant mu band (8-12Hz) de-synchronization , beginning approximately 0.5s before the imagination onset cue, and remaining the entire task duration. After the cue onset, we observed beta-band (15-30Hz) bursts (ERD) with contralateral dominance and mu band bursts (ERS) with ipsilateral dominance. Our results confirm that some of the neural substrates of motor imagery indeed overlap with those of motor execution. Specifically, the EEG dynamics and cortical localization of motor imagery readiness potential will help in constructing more efficient BMI decoders.	P1-1-253 ゲーム性を有するBMI訓練システムの開発 Development of videogame training system for motor-image BMI ○椿田紘久1, 恩田壮恭1, 石山敦士1, 小野弓絵2 ○Hirohisa Tsubakida1, Masanori Onda1, Atsushi Ishiyama1, Yumie Ono2 早稲田大学　先進理工学部　電気・情報生命工学科1, 明治大学理工学部電気電子生命学科2 School of Advanced Science and Engineering, Waseda Univ, Tokyo1, Dept. of Elec. and Bioinformatics, Sch. of Sci. and Tech., Meiji Univ., Kanagawa, Japan2 Our purpose is to control external equipment with change in Electroencephalography (EEG) induced by motor imagery (motor-imagery brain machine interface: MI-BMI). Considering that training is necessary to successfully control MI-BMI, adding an element of entertainment to the training phase may enhance the motivation of users and achieve better controllability. We therefore developed motor imagery training system having videogame-like interfaces and investigated its effectiveness. Five healthy young-adult subjects participated in this study. EEG signals were recorded at 128Hz from 15 electrodes placed at Fp1, Fp2, F3, Fz, F4, T3, C3, Cz, C4, T4, P3, Pz, P4, O1, and O2 according to the International 10-20 System. We applied spatial filtering method to raw EEG signals to remove artifacts, and the preprocessed EEG signals were further subjected to Fast Fourier transform to determine feature vectors. Support vector machine classified the feature vectors into three types of motor imagery states (left hand, right hand and rest). Subjects had trained the three types of motor imagery tasks for five days (20~25 minutes per day) with the training system. This training system gave subjects visual feedback on the PC screen though a simple game. The number of successful trials in the game was shown as a score at the end of the game. We recorded changes of the scores along the trial days for every subject. Our results demonstrated that three of five subjects improved their scores through the 5 days of training. The average score increased from 0.33±0.22 to 6.7±1.2 in these subjects (15 for 100% success). The result demonstrates the effectiveness of the training system which may allow possible users to quickly and effectively accustom to MI-BMI system.
P1-1-254 適合ウェーブレットと粒子群最適化を用いた聴覚事象関連電位の分類に関する検討 Classification of auditory event-related potentials using adapted wavelets and Particle Swarm Optimization ○ゴンザレスアレハンドロ1, 南部功夫1, 穂刈治英1, 和田安弘1 ○Alejandro Gonzalez1, Isao Nambu1, Haruhide Hokari1, Yasuhiro Wada1 長岡技術科学大学大学院　工学研究科　電気電子情報工学専攻1 Department of Electrical, Electronics and Information Engineering, Nagaoka University of Technology, Nagaoka, Japan1 In the present work we propose a method to classify auditory event-related potentials (ERPs) using the Discrete Wavelet Transform (DWT) and Particle Swarm Optimization (PSO) and evaluate it on measured data. The classification of ERPs is of great importance in the development of Brain-machine Interfaces (BMIs) and in this work we focus on an auditory BMI system in which the direction of attention of the user is estimated by detecting the P300 ERP. This ERP is elicited through an oddball paradigm with a virtual acoustic source and measured using EEG, and its detection relies on the appropriate feature extraction from the EEG signals and the proper tuning of a classifier. The P300 is a voltage deflection that is localized in both time and frequency and we propose to use the DWT to exploit this property and extract the most relevant features for classification. Furthermore, to account for differences between users, an adaptive approach that adjusts the wavelet basis to each user is proposed. We use PSO to search the wavelet basis that maximizes classification accuracy on training data. The DWT is represented using the Lifting Scheme, a representation that imposes fewer restrictions than the classical one. Additionally, the algorithm searches the best EEG channel combination to reduce the number of features. Support Vector Machines with a Gaussian kernel were used for classification and its parameters were also tuned with the PSO algorithm. We performed simulations using measured data of 3 subjects and assessed the average classification performance of the proposed algorithm using 5-fold cross-validation. We tested the algorithm on both single-trial and averaged data obtaining an average accuracy of 75.5% and 87.6%, respectively. The adapted wavelet bases resembled the P300 ERP in various cases, suggesting the proposed method can adapt to the ERP of the user. Additionally, the proposed method could attain a performance similar to previous reports using fewer channels.

上部に戻る

前に戻る