神経デコーディング
Neural decoding
O1-8-5-1
MEGを用いたリアルタイム義手制御
Real-time prosthetic arm control using MEG signals

○福間良平1,2, 柳澤琢史1,3,4, 平田雅之3, 菅田陽怜3, 松下光次郎3, 加藤龍5, 關達也5, 貴島晴彦3, 横井浩史5, 神谷之康1,2, 吉峰俊樹3
○Ryohei Fukuma1,2, Takufumi Yanagisawa1,3,4, Masayuki Hirata3, Hisato Sugata3, Kojiro Matsushita3, Ryu Kato5, Tatsuya Seki5, Haruhiko Kishima3, Hiroshi Yokoi5, Yukiyasu Kamitani1,2, Toshiki Yoshimine3
(株)国際電気通信基礎技術研究所 脳情報研究所1, 奈良先端科学技術大学院大学2, 大阪大学大学院医学系研究科脳神経外科3, 大阪大学大学院医学系研究科神経機能診断学4, 電気通信大学知能機械工学専攻5
ATR Computational Neuroscience Laboratories, Kyoto1, Nara Institute of Science and Technology, Nara2, Department of Neurosurgery, Osaka University Medical School, Osaka3, Division of Functional Diagnostic Science, Osaka University Medical School, Osaka4, Department of Mechanical Engineering and Intelligent Systems, The University of Electro-Communications, Tokyo5

Objective: Magnetoencephalography (MEG) has been shown to detect signals to infer human motor information with excellent time resolution. Combined with on-line analysis, MEG signals can be used for brain machine interfaces (BMI), which may assist evaluation and rehabilitation of patients with motor disorders. Here, we present a realtime MEG system that uses MEG signals to control a prosthetic arm.
Methods: MEG signals from a 160-channel Yokogawa MEG system with a real time output port were analyzed by an online system. We measured the signals of normal volunteers and amyotrophic lateral sclerosis (ALS) patients, while they asked to attempt to perform grasping or opening movements with their hand with auditory and visual cues given every 5.5 s. The ALS patients could not move their hand properly due to their severe motor impairments. The performed movement types were predicted from time-averaged MEG signals of the sensors around the motor cortex using support vector machine (SVM). Movement onsets were inferred by a Gaussian process regression (GPR) model combined with SVM. The prosthetic arm was controlled at the inferred movement onset to perform the predicted movement.
Results: The movement type was predicted with approximately 70% precision even in ALS patients from MEG signals around the cue timing. Statistical analysis revealed that the movement-related signals around the motor area significantly contributed to the decoding. The GPR model accurately detected the movement onset without the information about onset cues. The movement type at the detected timing was decoded with approximately 70% precision by the SVM decoder.
O1-8-5-2
日本手話における個人の言語能力に対する解剖学的指標: VBM 研究
Anatomical signature for the linguistic proficiency of individuals in Japanese Sign Language: A VBM study

○犬伏知生1,2, 酒井邦嘉1,3
○Tomoo Inubushi1,2, Kuniyoshi L. Sakai1,3
東京大院・総合文化・相関基礎1, 日本学術振興会2, CREST, 日本科学技術振興機構3
Dept. of Basic Sci., Univ. of Tokyo, Tokyo, Japan1, JSPS, Tokyo, Japan2, CREST, JST, Tokyo, Japan3

A unique opportunity to reveal the anatomical signature for the linguistic proficiency is provided by Deaf individuals with the considerable variability of the proficiency in Japanese Sign Language (JSL). We have shown the commonality of brain activations during lexical, syntactic, and contextual decisions between spoken and sign languages. Here, we conducted a voxel-based morphometry (VBM) study with three language conditions: lexical (Lex), syntactic (Syn), and contextual decision conditions (Con), as well as a baseline repetition condition (R). Under each of the language conditions, a probe-detection task was performed, in which Deaf participants (N = 30) detected a lexical, syntactic, or contextual error in JSL sentences. Under the R, the participants detected the repetition of reversed video-taped images used in the linguistic conditions. The task performances of these conditions were evaluated with d´. According to multiple regression analyses, there was a significant positive correlation between d´ and gray matter (GM) volumes in localized regions (corrected p < 0.05), even when the ages of acquisition (AOAs) of JSL and Japanese were factored out. Between the d´ of the Lex and GM volumes, we observed a prominent correlation in the dorsolateral surface of the left precentral and postcentral gyri. This region corresponded to the "hand area" of the primary motor and somatosensory cortices, indicating the better acquisition of subtle and complex hand movements by signers. Between the d´ of the Syn and GM volumes, a significant correlation was observed in the right insula. Finally, we found a strong correlation between the d´ of the Con and GM volumes in the opercular part of the left inferior frontal gyrus, which was consistent with its pivotal role for sentence comprehension suggested in our previous studies. These results suggest anatomical specialization of these regions related to inter-individual differences of the linguistic proficiency in sign languages.
O1-8-5-3
Is it sensory or motor error that is carried by climbing fiber for driving cerebellar plasticity? Evaluation by a cerebellar neuronal network model in real-world robot control
○Ruben Pinzon Morales1, Yutaka Hirata1
Department of Computer Science, Chubu University1

It has been shown that the activities of climbing fiber (cf) drive the parallel fiber-Purkinje cell synaptic plasticity in the cerebellum, and this synaptic mechanism is assumed to be a basis of motor learning. However, it is still unclear what kind of information is encoded in the cf activities. There are two major hypotheses; sensory error (SE) or motor error (ME) information. In the present study, to assess the feasibility of SE and ME, we configured a neuronal network model of the cerebellum, and tested it in real-time control of a two-wheeled balancing robot. The model is implemented in LabVIEW, consisting of 7 mossy fibers (mfs), 755 granular cells, 5 Golgi cells, 15 basket/stellate cells, 1 Prukinje (Pk) cell, and 1 cf. The 7 mfs transport desired motions of body tilt and wheel angle, control errors of body tilt and wheel angle, and efference copy of the motor command that drives the 2 wheels. For the cf activity, SE was computed as the difference between the desired and actual motions, whereas ME was the output of a feed-back controller that works in parallel with the cerebellar model. The performance of the robot control was evaluated by using 100 repetitions of a trapezoidal desired motion (max. speed 4 cm/s, cycle 10 s) for wheels with zero body tilt angle. Five types of cf were tested: #1 SE in position, #2 SE in velocity, #3 SE in both velocity and position, #4 ME, and #5 ME with SE in position. Results showed that #1 and #3 were able to produce proper synaptic plasticity in the model to keep controlling the robot, but exhibiting large control errors due to lagged Pk cell activity. Contrary, cf carrying #2 failed to govern the robot. Cf with #4 yielded better performance with lower control errors than #1 and #3. Finally, #5 achieved the best performance of all. Thus either SE or ME alone are feasible as cf activity to drive adequate synaptic changes in the cerebellar model. However, ME outperformed SE, and a combination of the two results in the best.
O1-8-5-4
ニューロフィードバック訓練中の脳直流電位成分(SCP)のコヒーレンス解析:頭皮上と硬膜下記録の相関
Correlation between scalp-recorded and subdural slow cortical potentials: direct comparison during neuro-feedback training

○文室知之1, 松本理器1, 松橋眞生2, 宇佐美清英1, 下竹昭寛1, 國枝武治3, 高橋良輔1, 池田昭夫1
○Tomoyuki Fumuro1, Riki Matsumoto1, Masao Matsuhashi2, Kiyohide Usami1, Akihiro Shimotake1, Takeharu Kunieda3, Ryosuke Takahashi1, Akio Ikeda1
京都大学大学院 医学研究科 臨床神経学1, 京都大学大学院 医学研究科 脳機能総合研究センター2, 京都大学大学院 医学研究科 脳神経外科学3
Dept Neurology, Kyoto Univ, Kyoto1, Human Brain Research Center, Kyoto Univ, Kyoto2, Dept Neurosurgery, Kyoto Univ, Kyoto3

Recently, neuro-feedback (NFB) training by self-regulation of slow cortical potentials (SCPs) recorded at scalp vertex has been applied for seizure suppression in patients with refractory epilepsy. However, it is not clear whether the whole cortices or parts of the cortices contribute to the scalp-recorded SCPs. It is also uncertain whether scalp-recorded SCPs may contain slow component of galvanic-skin response. To clarify the two questions, we evaluated the correlation of SCPs between scalp and subdural recording by means of the coherence analysis. In 6 patients with refractory partial epilepsy while invasive recording was done before epilepsy surgery, scalp and subdural SCPs were recorded simultaneously during NFB training by means of DC-EEG machine (IRB#C533). SCPs in C1 or C2 (contralateral to the subdural electrodes) was employed as a reference to the coherence analysis. As a result, for scalp-recorded SCPs, SCPs on the primary sensorimotor cortices had the highest correlation, than those of lateral (P=0.002) and basal temporal (P=0.001) cortices. It is most likely that scalp-recorded SCPs from the scalp vertex area exclusively could reflect SCPs of the cortices of the lateral convexity close to the vertex, and thus that scalp-recorded SCPs do not contain the galvanic skin response.
上部に戻る 前に戻る