大鼠初级运动皮层的神经信号分析与解码研究
|
摘要
脑机接口(Brain-machine Interface, BMI)是不依赖于常规的脊髓/外周神经肌肉系统在脑与外部设备间创建的直接连接通路。脑机接口系统的主要任务是采集大脑神经信号,分析和解码神经信号预测运动参数并将其转化为外围设备控制命令。本课题以大鼠为研究对象,建立了一套针对包括初级运动皮层局部场电位信号和锋电位信号的神经信号的特征提取、特性分析和信息解码的方法。
本课题首先研究了两种神经信号的特征提取方法。针对局部场电位信号主要研究采用Morlet小波变换提取多通道、多频段的时频域特征信息；针对锋电位信号建立了基于谱聚类算法和朴素贝叶斯分类器的锋电位分类方法,结果表明该锋电位分类方法具有较好的精度、抗噪性和效率。
在提取出神经信号的特征后,本课题接着对两种信号特征分别从定性和定量的角度作了分析。定性分析方面,本课题使用等距映射流行学习算法对高维度的神经信号特征做了三维可视化分析,结果表明神经信号在运动过程中具有规律性的变化模式,并且局部场电位信号和锋电位信号的变化模式不相同；定量分析方面,本课题通过估计两种神经信号高维特征与运动参数的互信息分析了神经信号与运动参数的相关性,分析结果表明两种神经信号均包含了运动相关的信息,且在通道较少的情况下联合两种信号可以获得比单独采用其中一种神经信号获得更多的运动相关信息。
最后,为了从神经信号中解析和重建出运动信息,本课题研究了神经信息解码方法。针对多通道多频段神经信号特征带来的高输入维度问题,本课题建立了基于Boosting思想的特征子集筛选算法和基于偏最小二乘的降维算法两种神经信息属性约简策略,提高了解码模型的实时性和泛化能力。对于解码算法,本课题分别实现和比较了偏最小二乘线性解码算法及核偏最小二乘算法非线性解码算法；在进一步分析线性简单模型和非线性复杂模型的利弊后,本课题提出了兼具高运算效率和高解码精度的两阶段解码模型。本课题还实现了局部场电位信号和锋电位信号的联合解码,实验结果表明联合两种信号提高了解码精度和解码模型的稳定性,解决了单一神经信号的信息不完备性对系统整体解码性能的影响。
Brain-machine interface (BMI) provides a directly communication channel between the brain and man-made devices independent of peripheral nervous system. The main functions of a BMI system include neural signals recording, neural signals analysis and decoding to predict the movement parameters and convert them into controlling commands of man-made devices. Research presented in this study provides a set of methods and schemes on feature extraction, analysis and information decoding of local field potential (LFP) signal and spike signal, using rat as the experiment object.
Firstly, as a important preprocessing procedure feature extraction of LFP and spike were studied. For LFP, Morlet wavelet transform was applied to extract the multi-channel and multi-band time-frequency domain features. For spike, spike-sorting system based on Spectral Clustering and Naive Bayesian Classification algorithms was built. Experiment results show this systerm has properties of high precision, noise robustness and high computational efficiency.
After feature extraction of neural signals, qualitative and quantitative analysis of the features were researched secondly. For qualitative analysis, a manifold algorithm namely Isomap was used to reduce the dimension of the high dimensional neural signal and visualize the variety pattern of the neural signals in a 3-D space. The visualization results show that both types of neural signals have regular variety patterns during the experiment movement but the patterns of LFP and spike are different. For quantitative analysis, mutual informations between the two types of high dimensional neural signal features and the motion parameter were estimated to analyze the relativity between neural signals and the movement. Relativity analysis results show that both LFP and spike contain abundant information about the movement and combining LFP and spike can get more useful information than taking each of the two types of signal alone on condition of small number of channels.
At last, neural information decoding system were studied in order to decode and reconstruct the movement related information. Two kinds of neural feature reduction strategies, including Boosting-based sub-feature space selection algorithm and partial least square (PLS)-based dimension reduction algorithm, were applied to enhance the real-time and generalization capability of the decoding system. For the decoding algorithms, partial least square regression linear algorithm and kernel partial least square (KPLS) regression non-linear algorithm were studied and compared. After studying and comparing the advantages and disadvantages of linear and non-linear decoding algorithms, a Two Stage Model was built to achieve both the high computational efficient advantage of linear algorithms and the high decoding accuracy advantage of non-linear algorithms. Furthermore, the decoding power of combining LFP and spike was also analyzed in this study. Decoding results indicate that combining these two types of neural signals can improve both decoding accuracy and stability of the decoding model.
本课题首先研究了两种神经信号的特征提取方法。针对局部场电位信号主要研究采用Morlet小波变换提取多通道、多频段的时频域特征信息；针对锋电位信号建立了基于谱聚类算法和朴素贝叶斯分类器的锋电位分类方法,结果表明该锋电位分类方法具有较好的精度、抗噪性和效率。
在提取出神经信号的特征后,本课题接着对两种信号特征分别从定性和定量的角度作了分析。定性分析方面,本课题使用等距映射流行学习算法对高维度的神经信号特征做了三维可视化分析,结果表明神经信号在运动过程中具有规律性的变化模式,并且局部场电位信号和锋电位信号的变化模式不相同；定量分析方面,本课题通过估计两种神经信号高维特征与运动参数的互信息分析了神经信号与运动参数的相关性,分析结果表明两种神经信号均包含了运动相关的信息,且在通道较少的情况下联合两种信号可以获得比单独采用其中一种神经信号获得更多的运动相关信息。
最后,为了从神经信号中解析和重建出运动信息,本课题研究了神经信息解码方法。针对多通道多频段神经信号特征带来的高输入维度问题,本课题建立了基于Boosting思想的特征子集筛选算法和基于偏最小二乘的降维算法两种神经信息属性约简策略,提高了解码模型的实时性和泛化能力。对于解码算法,本课题分别实现和比较了偏最小二乘线性解码算法及核偏最小二乘算法非线性解码算法；在进一步分析线性简单模型和非线性复杂模型的利弊后,本课题提出了兼具高运算效率和高解码精度的两阶段解码模型。本课题还实现了局部场电位信号和锋电位信号的联合解码,实验结果表明联合两种信号提高了解码精度和解码模型的稳定性,解决了单一神经信号的信息不完备性对系统整体解码性能的影响。
Brain-machine interface (BMI) provides a directly communication channel between the brain and man-made devices independent of peripheral nervous system. The main functions of a BMI system include neural signals recording, neural signals analysis and decoding to predict the movement parameters and convert them into controlling commands of man-made devices. Research presented in this study provides a set of methods and schemes on feature extraction, analysis and information decoding of local field potential (LFP) signal and spike signal, using rat as the experiment object.
Firstly, as a important preprocessing procedure feature extraction of LFP and spike were studied. For LFP, Morlet wavelet transform was applied to extract the multi-channel and multi-band time-frequency domain features. For spike, spike-sorting system based on Spectral Clustering and Naive Bayesian Classification algorithms was built. Experiment results show this systerm has properties of high precision, noise robustness and high computational efficiency.
After feature extraction of neural signals, qualitative and quantitative analysis of the features were researched secondly. For qualitative analysis, a manifold algorithm namely Isomap was used to reduce the dimension of the high dimensional neural signal and visualize the variety pattern of the neural signals in a 3-D space. The visualization results show that both types of neural signals have regular variety patterns during the experiment movement but the patterns of LFP and spike are different. For quantitative analysis, mutual informations between the two types of high dimensional neural signal features and the motion parameter were estimated to analyze the relativity between neural signals and the movement. Relativity analysis results show that both LFP and spike contain abundant information about the movement and combining LFP and spike can get more useful information than taking each of the two types of signal alone on condition of small number of channels.
At last, neural information decoding system were studied in order to decode and reconstruct the movement related information. Two kinds of neural feature reduction strategies, including Boosting-based sub-feature space selection algorithm and partial least square (PLS)-based dimension reduction algorithm, were applied to enhance the real-time and generalization capability of the decoding system. For the decoding algorithms, partial least square regression linear algorithm and kernel partial least square (KPLS) regression non-linear algorithm were studied and compared. After studying and comparing the advantages and disadvantages of linear and non-linear decoding algorithms, a Two Stage Model was built to achieve both the high computational efficient advantage of linear algorithms and the high decoding accuracy advantage of non-linear algorithms. Furthermore, the decoding power of combining LFP and spike was also analyzed in this study. Decoding results indicate that combining these two types of neural signals can improve both decoding accuracy and stability of the decoding model.
引文
[1]Chapin, J. K., K. A. Moxon, et al. Real-time control of a robot arm using simultaneously recorded neurons in the motor cortex[J]. Nature Neuroscience,1999, 1999(2):664-670.
[2]Serruya, M. D., N. G. Hatsopoulos, et al. Brain-Machine interface:Instant neural control of a movement signal[J]. Nature,2002,416(6877):141-142.
[3]Carmena, J. M., M. A. Lebedev, et al. Learning to control a Brain-Machine interface for reaching and grasping by primates[J]. PLoS Biology,2003,1(2):e42.
[4]Mollazadeh, M., V. Aggarwal, et al. Spectral modulation of LFP activity in M1 during dexterous finger movements[C]. Conf Proc IEEE Eng Med Biol Soc.,2008, 2008:5314-5317.
[5]Lebedev, M. A. and M. A. L. Nicolelis. Brain-machine interfaces:past, present and future[J]. TRENDS in Neurosciences,2006,29(9):536-546.
[6]Wessberg, J., C. R. Stambaugh, et al. Real-time prediction of hand trajectory by ensembles of cortical neurons in primates[J]. Nature,2000,408(6810):361-365.
[7]Hochberg, L. R., M. D. Serruya, et al. Neuronal ensemble control of prosthetic devices by a human with tetraplegia[J]. Nature,2006,442(7099):164-171.
[8]Zhuang, J., W. Truccolo, et al. Decoding 3-D Reach and Grasp Kinematics From High-Frequency Local Field Potentials in Primate Primary Motor Cortex[J]. IEEE Transactions on Biomedical Engineering,2010,57(7):1774-1784.
[9]A.K. Bansal, C.E. Vargas-Irwin, W. Truccolo, and J.P. Donoghue. Relationships among low-frequency local field potentials, spiking activity, and 3-D reach and grasp kinematics in primary motor and ventral premotor cortices [J]. Journal of Neurophysiology,2011,105(4):1603-1619.
[10]M. Lewicki. A review of methods for spike sorting:the detection and classification of neural action potentials[J]. Network:Computation in Neural Systems, 1998,9(4):53-78.
[11]M. Nicolelis. Actions from thoughts[J]. Nature,2001,409(6818):403-407.
[12]Mehring, C., J. Rickert, et al. Inference of hand movements from local field potentials in monkey motor cortex[J]. nature neuroscience,2003,6(12):1253-1254.
[13]Mollazadeh, M., V. Aggarwal, et al. Coherency between spike and LFP activity in M1 during hand movements[C]. IEEE Eng. Med. Biol. Soci.,2009,2009:506-509.
[14]Georgopoulos, A.P., A.B. Schwartz, and R.E. Kettner. Neuronal population coding of movement direction [J]. Science,1986,233(4771):1416-1419.
[15]L. Paninski, M. R. Fellows, N. G. Hatsopoulos et al.. Spatiotemporal tuning of motor cortical neurons for hand position and velocity[J]. Journal of Neurophysiology, 2004,91(1):515-532.
[16]Wu, W., M.J. Black, D. Munford, Y. Gao, E. Bienenstock, and J.P. Donoghue, Modeling and decoding motor cortical activity using a switching Kalman filter[J]. IEEE Transactions on Biomedical Engineering,2004,51:933-942.
[17]Gao, Y., et al. A quantitative comparison of linear and non-linear models of motor cortical activity for the encoding and decoding of arm motions[C].1st International IEEE EMBS Conference on Neural Engineering,2003,2003:189-192.
[18]Andrieu, C., et al., Particle methods for change detection, system identifcation, and control[J]. Proceed-ings of the IEEE,2004,92(3):423-438.
[19]B. Olson, J. Si, J. Hu et al.. Closed-loop cortical control of direction using support vector machines[J]. Neural Systems and Rehabilitation Engineering,IEEE Transactions on,2005,13(1),72-80.
[20]Cynthia A. Chestek, Aaron P. Batista, Gopal Santhanam et al..Single-Neuron Stability during Repeated Reaching in Macaque Premotor Cortex[J]. Journal of Neuroscience,2007,27(40),10742-10750.
[21]Mehring, C., J. Rickert, et al.. Inference of hand movements from local field potentials in monkey motor cortex[J]. nature neuroscience,2003,6(12):1253-1254.
[22]代建华,章怀坚,张韶岷et al大鼠运动皮层神经元集群锋电位时空模式解析[J].中国科学：C辑,1999,736-745.
[23]Stark, E. and M. Abeles. Predicting movement from multiunit activity[J]. Journal of Neuroscience,2007,27(31):8387-8394.
[24]Chao, Z., Y. Nagasaka, et al. Long-term asynchronous decoding of arm motion using electrocorticographic signals in monkeys [J]. Frontiers in Neuroengineering, 2010,3:3.
[25]D. Gabor. Theory of communication[J]. Journal of the Institution of Electrical Engineers,1946,93(26):429-457.
[26]胡广书.第2章短时傅里叶变换.现代信号处理教程[M].清华大学出版社,2004,48-50.
[27]胡广书.第9章小波变换基础.现代信号处理教程[M].清华大学出版社,2004,246-249.
[28]Ng, A., Jordan, M., and Weiss, Y.. On spectral clustering:analysis and an algorithm [J]. Advances in Neural Information Processing Systems,2002,14(14): 849-856.
[29]L Zelnik-Manor, P Perona. Self-Tuning Spectral Clustering [J]. Advances in neural information processing systems,2004,2:1601-1608.
[30]I Rish. An empirical study of the naive Bayes classifier[J]. IJCAI 2001 Workshop on Empirical Methods in Artificial Intelligence,2001,3(22):41-46.
[31]徐蓉,姜峰,姚鸿勋.流形学习概述[J].智能系统学报,2006,1(1)：44-51.
[32]Joshua B. Tenenbaum, Vin de Silva and John C. Langford. A Global Geometric Framework for Nonlinear Dimensionality Reduction[J]. Science,2000,290(5500): 2319-2323.
[33]Jack B Arnold. A Multidimensional Scaling Study of Semantic Distance[J]. Journal of Experimental Psychology Monograph,1971,90(2):349-372.
[34]Cormen, Thomas H., Leiserson, Charles E., Rivest, Ronald L., Stein, Clifford. "Section 24.3:Dijkstra's algorithm". Introduction to Algorithms (Second ed.)[M]. MIT Press and McGraw-Hill,2001,595-601.
[35]La The Vinh, Sungyoung Lee, Young-Tack Park and Brian J. d'Auriol. A novel feature selection method based on normalized mutual information [J]. Applied Intelligence,2011,2011:1-21.
[36]C.E. Shannon, A Mathematical Theory of Communication[J]. Bell System Technical Journal,1948,27:379-423,623-656.
[37]Rosenblatt, M.. Remarks on some nonparametric estimates of a density function[J]. Annals of Mathematical Statistics,1956,27:832-837.
[38]Parzen, E.. On estimation of a probability density function and mode[J]. Annals of Mathematical Statistics,1962,33:1065-1076.
[39]David W. Scott. Scott's rule[J].Wiley Interdisciplinary Reviews:Computational Statistics,2010,2(4):497-502.
[40]Robert E. Schapire. The Strength of Weak Learnability[J]. Machine Learning, 1990,5(2):197-227.
[41]Wold S, Albano C, Dunn M, et al. Pattern regression finding and using regularities in multivariate data[C]. In Martens J In Proc IUFOST Conf' Food Research and Data Analysis. London:Analysis Applied Science Publication,1983,1983:147-188.
[42]王惠文.PLS回归对多变量信息的综合筛选作用分析[J].数据统计与管理,1998,17(4)：46-49.
[43]Rosipal R,Trejo L. Kernel Partial Least Squares Regression in Reproducing Kernel Hilbert Space[J]. Journal of Machine Learning Research,2001,2:97-123.
[44]Yifeng B, Jian X, Long Y. Kernel partial least-squares regression[C]. International Joint Conference on Neural Networks,2006,2006:1231-1238.
[45]J. Dai, X. Liu, S. Zhang, H. Zhang, Q. Xu, W. Chen, and X. Zheng. Continuous Neural Decoding Method Based on General Regression Neural Network[J]. International Journal of Digital Content Technology and its Applications,2010,4(8): 216-221.
[2]Serruya, M. D., N. G. Hatsopoulos, et al. Brain-Machine interface:Instant neural control of a movement signal[J]. Nature,2002,416(6877):141-142.
[3]Carmena, J. M., M. A. Lebedev, et al. Learning to control a Brain-Machine interface for reaching and grasping by primates[J]. PLoS Biology,2003,1(2):e42.
[4]Mollazadeh, M., V. Aggarwal, et al. Spectral modulation of LFP activity in M1 during dexterous finger movements[C]. Conf Proc IEEE Eng Med Biol Soc.,2008, 2008:5314-5317.
[5]Lebedev, M. A. and M. A. L. Nicolelis. Brain-machine interfaces:past, present and future[J]. TRENDS in Neurosciences,2006,29(9):536-546.
[6]Wessberg, J., C. R. Stambaugh, et al. Real-time prediction of hand trajectory by ensembles of cortical neurons in primates[J]. Nature,2000,408(6810):361-365.
[7]Hochberg, L. R., M. D. Serruya, et al. Neuronal ensemble control of prosthetic devices by a human with tetraplegia[J]. Nature,2006,442(7099):164-171.
[8]Zhuang, J., W. Truccolo, et al. Decoding 3-D Reach and Grasp Kinematics From High-Frequency Local Field Potentials in Primate Primary Motor Cortex[J]. IEEE Transactions on Biomedical Engineering,2010,57(7):1774-1784.
[9]A.K. Bansal, C.E. Vargas-Irwin, W. Truccolo, and J.P. Donoghue. Relationships among low-frequency local field potentials, spiking activity, and 3-D reach and grasp kinematics in primary motor and ventral premotor cortices [J]. Journal of Neurophysiology,2011,105(4):1603-1619.
[10]M. Lewicki. A review of methods for spike sorting:the detection and classification of neural action potentials[J]. Network:Computation in Neural Systems, 1998,9(4):53-78.
[11]M. Nicolelis. Actions from thoughts[J]. Nature,2001,409(6818):403-407.
[12]Mehring, C., J. Rickert, et al. Inference of hand movements from local field potentials in monkey motor cortex[J]. nature neuroscience,2003,6(12):1253-1254.
[13]Mollazadeh, M., V. Aggarwal, et al. Coherency between spike and LFP activity in M1 during hand movements[C]. IEEE Eng. Med. Biol. Soci.,2009,2009:506-509.
[14]Georgopoulos, A.P., A.B. Schwartz, and R.E. Kettner. Neuronal population coding of movement direction [J]. Science,1986,233(4771):1416-1419.
[15]L. Paninski, M. R. Fellows, N. G. Hatsopoulos et al.. Spatiotemporal tuning of motor cortical neurons for hand position and velocity[J]. Journal of Neurophysiology, 2004,91(1):515-532.
[16]Wu, W., M.J. Black, D. Munford, Y. Gao, E. Bienenstock, and J.P. Donoghue, Modeling and decoding motor cortical activity using a switching Kalman filter[J]. IEEE Transactions on Biomedical Engineering,2004,51:933-942.
[17]Gao, Y., et al. A quantitative comparison of linear and non-linear models of motor cortical activity for the encoding and decoding of arm motions[C].1st International IEEE EMBS Conference on Neural Engineering,2003,2003:189-192.
[18]Andrieu, C., et al., Particle methods for change detection, system identifcation, and control[J]. Proceed-ings of the IEEE,2004,92(3):423-438.
[19]B. Olson, J. Si, J. Hu et al.. Closed-loop cortical control of direction using support vector machines[J]. Neural Systems and Rehabilitation Engineering,IEEE Transactions on,2005,13(1),72-80.
[20]Cynthia A. Chestek, Aaron P. Batista, Gopal Santhanam et al..Single-Neuron Stability during Repeated Reaching in Macaque Premotor Cortex[J]. Journal of Neuroscience,2007,27(40),10742-10750.
[21]Mehring, C., J. Rickert, et al.. Inference of hand movements from local field potentials in monkey motor cortex[J]. nature neuroscience,2003,6(12):1253-1254.
[22]代建华,章怀坚,张韶岷et al大鼠运动皮层神经元集群锋电位时空模式解析[J].中国科学：C辑,1999,736-745.
[23]Stark, E. and M. Abeles. Predicting movement from multiunit activity[J]. Journal of Neuroscience,2007,27(31):8387-8394.
[24]Chao, Z., Y. Nagasaka, et al. Long-term asynchronous decoding of arm motion using electrocorticographic signals in monkeys [J]. Frontiers in Neuroengineering, 2010,3:3.
[25]D. Gabor. Theory of communication[J]. Journal of the Institution of Electrical Engineers,1946,93(26):429-457.
[26]胡广书.第2章短时傅里叶变换.现代信号处理教程[M].清华大学出版社,2004,48-50.
[27]胡广书.第9章小波变换基础.现代信号处理教程[M].清华大学出版社,2004,246-249.
[28]Ng, A., Jordan, M., and Weiss, Y.. On spectral clustering:analysis and an algorithm [J]. Advances in Neural Information Processing Systems,2002,14(14): 849-856.
[29]L Zelnik-Manor, P Perona. Self-Tuning Spectral Clustering [J]. Advances in neural information processing systems,2004,2:1601-1608.
[30]I Rish. An empirical study of the naive Bayes classifier[J]. IJCAI 2001 Workshop on Empirical Methods in Artificial Intelligence,2001,3(22):41-46.
[31]徐蓉,姜峰,姚鸿勋.流形学习概述[J].智能系统学报,2006,1(1)：44-51.
[32]Joshua B. Tenenbaum, Vin de Silva and John C. Langford. A Global Geometric Framework for Nonlinear Dimensionality Reduction[J]. Science,2000,290(5500): 2319-2323.
[33]Jack B Arnold. A Multidimensional Scaling Study of Semantic Distance[J]. Journal of Experimental Psychology Monograph,1971,90(2):349-372.
[34]Cormen, Thomas H., Leiserson, Charles E., Rivest, Ronald L., Stein, Clifford. "Section 24.3:Dijkstra's algorithm". Introduction to Algorithms (Second ed.)[M]. MIT Press and McGraw-Hill,2001,595-601.
[35]La The Vinh, Sungyoung Lee, Young-Tack Park and Brian J. d'Auriol. A novel feature selection method based on normalized mutual information [J]. Applied Intelligence,2011,2011:1-21.
[36]C.E. Shannon, A Mathematical Theory of Communication[J]. Bell System Technical Journal,1948,27:379-423,623-656.
[37]Rosenblatt, M.. Remarks on some nonparametric estimates of a density function[J]. Annals of Mathematical Statistics,1956,27:832-837.
[38]Parzen, E.. On estimation of a probability density function and mode[J]. Annals of Mathematical Statistics,1962,33:1065-1076.
[39]David W. Scott. Scott's rule[J].Wiley Interdisciplinary Reviews:Computational Statistics,2010,2(4):497-502.
[40]Robert E. Schapire. The Strength of Weak Learnability[J]. Machine Learning, 1990,5(2):197-227.
[41]Wold S, Albano C, Dunn M, et al. Pattern regression finding and using regularities in multivariate data[C]. In Martens J In Proc IUFOST Conf' Food Research and Data Analysis. London:Analysis Applied Science Publication,1983,1983:147-188.
[42]王惠文.PLS回归对多变量信息的综合筛选作用分析[J].数据统计与管理,1998,17(4)：46-49.
[43]Rosipal R,Trejo L. Kernel Partial Least Squares Regression in Reproducing Kernel Hilbert Space[J]. Journal of Machine Learning Research,2001,2:97-123.
[44]Yifeng B, Jian X, Long Y. Kernel partial least-squares regression[C]. International Joint Conference on Neural Networks,2006,2006:1231-1238.
[45]J. Dai, X. Liu, S. Zhang, H. Zhang, Q. Xu, W. Chen, and X. Zheng. Continuous Neural Decoding Method Based on General Regression Neural Network[J]. International Journal of Digital Content Technology and its Applications,2010,4(8): 216-221.