神経系の大規模シミュレーションとハードウェア実現
Large scale simulation and hardware implementation of neural systems
O3-9-6-1
ギャップジャンクションが皮質回路の発火伝搬に与える影響の数値解析
Numerical analysis of the effect of gap junctions on firing propagation in a cortical feedforward network

○篠崎隆志1, 成瀬康1, 梅原広明1
○Takashi Shinozaki1, Yasushi Naruse1, Hiroaki Umehara1
NICT・脳情報1
Brain ICT Lab, NICT, Hyogo1

Gap junctions are electrical couplings between two cells connected by a binding protein, connexin. Theoretically, the role of gap junctions has mainly been investigated with respect to synchronization between electrically-coupled neurons (Moortgat et al., 2000; Pfeuty et al., 2003) and with respect to interactions between gap junctions and synaptic connections (Levis & Rinzel, 2003; Nomura et al., 2003). Recent electrophysiological studies have reported that a large number of inhibitory neurons in the mammalian cortex are mutually connected by gap junctions (Galarreta & Hestrin, 1999; Gibson et al., 1999; Fukuda et al., 2006), and synchronization of gap junctions, spread over several hundred microns (Beierlein et al., 2000), suggested a strong effect on the dynamics of the cortical network. Although most numerical simulations of firing propagation assume cortical circuits, the effect of gap junctions has not been examined systematically. In this study, we perform numerical simulations using biologically plausible parameters (Amitai et al., 2002; Levy & Reyes, 2012) and neuron model (Izhikevich, 2003) to clarify this effect with a synfire chain regime. The results demonstrate that gap junctions switch the temporally-uniform firing in a layer to temporally-clustered firing in subsequent layers, enhancing of the synfire chain propagation in the feedforward network. Quantitative analyses of the quality of the firing propagation show that the enhancement of gap junctions is the more effective in earlier layers and, moreover, too many gap junctions hinder the signal propagation. Because gap junctions are often modulated in physiological conditions (Mills & Massey, 1995; Bloomfield & Volgyi, 2004; Hipp et al., 2011), we speculate that gap junctions could be related to a gating function of population firing in lower sensory cortices.
O3-9-6-2
網膜神経節細胞のスパイク活動を再構築する実時間エミュレータ
A real-time emulation system for reconstructing spike activities of retinal ganglion cells

○奥野弘嗣1, 長谷川潤1, 八木哲也1
○Hirotsugu Okuno1, Jun Hasegawa1, Tetsuya Yagi1
大阪大学大学院 工学研究科1
Graduate School of Engineering, Osaka University, Osaka, Japan1

In order to discuss the functional roles of visual neurons in the retina, responses of a group of retinal neurons in natural visual environments should be investigated. In this study, we developed an emulation system for real-time reconstruction of neural activities of retinal neurons with physiologically feasible spatio-temporal properties. The system was implemented with an analog-digital-mixed image sensor system, which is mainly composed of resistive networks and a field-programmable gate array (FPGA). The system reconstructs the activities of 128 x 128 neurons in each neural layer of the retina at 200 frames per second. The spatial smoothing function mediated by the gap junctions of photoreceptors and those of horizontal cells were emulated by the first and the second layer of the resistive networks, respectively. The width of the smoothing filter can be easily controlled by externally applied voltage to the resisters because the resister used here is composed of transistors. The smoothed images are fed into the FPGA after digitized by an 8-bit ADC. In the FPGA, the center-surround antagonistic spatial property of bipolar cells is obtained by the difference of the output of the two resistive networks. Applying physiologically feasible temporal filters to this signal yields simulated responses of bipolar cells in the on-center and off-center pathways. The simulated response of a type of amacrine cell was computed by applying additional spatio-temporal filters to the simulated bipolar cell response. Employing the simulated responses of bipolar and amacrine cells, this system outputs spike trains generated by 4 types of simulated ganglion cells: sustained and transient cells in both the on-center and off-center pathways.We showed that the system can be used to visualize the complex nature of neural images formed by retinal neurons in dynamic contexts. The system could be a powerful platform to discuss and understand the functional roles of retinal neurons.
O3-9-6-3
胎児シミュレーションによる自発運動パターンの身体表象発達への寄与
How spontaneous movement patterns shape body representation in fetus simulation with spiking neural networks

○山田康智1,2, 藤井敬子3, 國吉康夫1
○Yasunori Yamada1,2, Keiko Fujii3, Yasuo Kuniyoshi1
東京大院・情理1, 学振・特別研究員(DC1)2, 東京大院・学府3
Grad. School of Info. Sci. & Tech., The Univ. of Tokyo, Japan1, JSPS research fellow2, Grad. School of Interdisciplinary Information Studies, The Univ. of Tokyo, Japan3

Converging developmental studies have emphasized the significance of learning through spontaneous movements from as early as the fetal period for nervous and cognitive development. In the somatosensory system, these spontaneous movements play an important role in guiding the organization of the nervous system, thereby forming the body representation for cognitive basis. However, the underlying mechanisms by which spontaneous movements form body representations are unclear. For example, are movement patterns important for the successful maturation of body representations? Here, we investigated how spontaneous movement patterns guide development of body representations, using computer simulations of fetus models with musculoskeletal body and spiking neural networks of leaky integrate-and-fire neurons with spike-timing-dependent synaptic plasticity. Organized networks emerged from random connectivities, and their structural and functional organizations were quantified using degrees of segregation and modularity. First, we examined the influences of neurobiological properties such as firing rates of input neurons, excitation/inhibition balance and background activity. We found that modular organizations which preserved the spatial structures of inputs emerged when input firing rates were high, excitation/inhibition balance was biologically realistic, and background activity was low. In fetus simulations, we showed that the above parameters and normal patterns of spontaneous movements, characterized by complexity and variability, guided body part-specific modular architectures. In contrast, we found that abnormal spontaneous movement patterns, which had low complexity and/or low variability, led to the formation of poorly organized body representations, with no body part-specific clusters and no modular architectures. Our results suggest that abnormalities in patterns of spontaneous movements may disrupt the normal development of body representations.
O3-9-6-4
PhysioDesignerを用いた画像ベースの高精度脳波・脳磁図シミュレーション
Image based realistic EEG/MEG simulation in PhysioDesigner

○岡秀樹1,2,3, 岩崎健一郎1, 浅井義之2, 野村泰伸3, 山口陽子1
○Hideki Oka1,2,3, Ken-ichiro Iwasaki1, Yoshiyuki Asai2, Taishin Nomura3, Yoko Yamaguchi1
(独)理化学研究所脳科学総合研究センター神経情報基盤センター1, 沖縄科学技術大学院大学2, 大阪大学3
NeuroInformatics Japan Center, RIKEN Science Institute, Saitama Japan1, Okinawa Institute of Science and Technology, Okinawa Japan2, Department of Mechanical Science and Bioengineering, Osaka University, Osaka Japan3

To do realistic high resolution EEG/MEG analysis, MRI image based EEG/MEG forward simulation system has been developed using PhysioDesigner1). Simulation model is described by modeling description language PHML1). PHML model is composed of four parts, morphology module, EEG module, MEG module and current source module. Morphology module gives a cubic simulation region in which brain structure is stored. EEG module solves a Poisson equation with inhomogeneous conductivity values extracted from brain MRI image and it gives EEG signals on 64ch electrode. MEG module solves a magnetic vector potential equation. It gives magnetic field values at the number of 306 sensors. Current source module gives information of rhythmic current source whose origin is a dipole. Now, it is composed of current strength module and frequency module. If necessary, it can be replaced by user defined rhythmic model.Brain conductivity can be extracted from MRI images of SPM decomposed grey, white, csf and skull images. Also, current dipole configuration must be defined. For the purpose, image processing tool Imageviewer has been developed. To the greyscale image region whose value is over the pre-determined the threshold value, definite conductivity value is set. Re-combining the four conductivity-set images, total conductivity file is generated and is used for Maxwell equations to get electric and magnetic field potentials. Dipole set function can set the positions of the dipole on the MRI image by mouse pointer operation or numerical values.To solve partial differential equations, finite element method is used. FEM can simulate by up to the same 1mm high resolution as MRI image. We adopt a script based solver FreeFem++2) whose script is parallelized to realize the resolution value. Obtained conductivity file is read from FreeFem++ and its values are assigned to each cubic point. (1) http://physiodesigner.org(2) http://www.frefem.org
上部に戻る 前に戻る