基于猴子M1区的腕部解码系统研究
发布时间：2018-08-19 18:25
【摘要】：脑机接口是指不依赖于常规的脊髓或者外周神经肌肉系统,在脑与外部设备之间建立一种新型的信息交流与控制通道,从而实现脑与外界的直接交互。目前,非人灵长类猕猴是脑机接口研究的主要动物模型。在以猕猴为模型的脑机接口研究中,关键难点是如何长期获得高质量的神经信号和运动信号,并通过构建解码算法实现神经信号对运动信号的解码预测。本文以非人灵长类猕猴为实验对象,对初级运动皮层(primary motor cortex, M1区)神经集群信号记录和腕部精细运动信号记录等方面的若干关键技术进行了探索性研究,构建了一个基于猕猴腕部运动的植入式脑机接口系统。通过对M1区神经信号的处理和分析,实现了对腕部精细运动的解码预测。本文主要研究内容包括： 首先,本文设计了一个腕部运动信号采集系统,应用于猴子腕部内翻、外翻、前展、后屈运动时运动参数的采集。该系统包括摇杆系统、固定系统、奖赏系统、PC (personal computer)机系统、下位机微控制器系统、摄像监控系统等。针对猴子腕部精细运动的特点,本文设计了简单的四方向center-out实验范式。经过任务分解训练,猴子可以较快地学会用腕部完成摇杆行为任务。该系统成功实现了猴子腕部运动信号的采集与记录。 第二,本文对猕猴大脑皮层神经集群记录的若干关键技术进行了探索性研究。设计了两种用于猴子头部固定的机械装置。比较分析了这两种装置的性能,发现八脚headpost和球头型head holder可以较好地用于猴子头部的固定。探索了两种接口犹他电极的植入技术,比较分析了两种接口犹他电极的优缺点。设计并改进了用于ICS-96接口犹他电极的固定基座,实现了犹他电极在M1区的精确埋植。基于本文建立的微电极阵列植入技术,能够从猴子运动皮层长期获得高质量的神经集群锋电位信号(spike)。 第三,同步采集获得猴子M1区的spike信号和腕部的运动信号,并定性分析了两者之间的相关性。首先,探索了神经集群发放模式与腕部运动方向之间的关系,发现神经集群的spike发放率在摇杆运动起始前后的变化最大。神经集群发放模式在不同摇杆方向上的差异性较大,这种差异性可用于不同摇杆“方向对”间的区分。其次,通过神经可视化算法分析神经信号与运动轨迹之间的关系,发现神经信号具有很强的内在规律性。降维后所得到的神经轨迹具有明显的可区分性,能够很好地反映实际运动规律。 最后,以神经元spike发放率为输入特征,通过解码算法实现了对猴子腕部运动方向、位置、速度的精确预测。此外,本文系统分析了影响解码效果的各种因素。本文分别选用K最近邻域算法(k-nearest neighbor algorithm,KNN)和支持向量机算法(support vector machine, SVM),实现了对猴子腕部运动方向的解码预测,预测正确率可以达到96%。此外,本文选用的卡尔曼滤波算法(Kalman Filter, KF)和广义回归神经网络算法(general regression neural network, GRNN)对位置和速度等运动参数进行解码分析。两种算法均取得了较好的解码效果。其中,GRNN算法对X、Y方向位置和速度的最高解码相关系数(correlation coefficient,CC)值可以达到0.9170±0.0458,0.8872±0.0778,0.8254±0.0798和0.8376±0.0915。 综上所述,利用猴子M1区记录的神经信号可以较好地预测出腕部的运动参数。本文建立的基于猴子M1区的腕部解码系统是一个成功的脑机接口系统。该植入式脑机接口系统为进一步研究大脑运动皮层的编解码规律,以及理解大脑控制运动的神经生物学机制奠定了基础。
[Abstract]:Brain-computer interface (BCI) refers to the establishment of a new type of communication and control channel between the brain and external devices, which is independent of the conventional spinal cord or peripheral neuromuscular system, so as to achieve direct interaction between the brain and the outside world. The key problem is how to obtain high-quality neural and motor signals for a long time, and how to decode and predict them by constructing decoding algorithms. Some key techniques such as motion signal recording have been explored and an implantable brain-computer interface system based on rhesus monkey wrist movement has been constructed.
Firstly, this paper designs a wrist motion signal acquisition system, which is used to collect the movement parameters of monkey wrist during varus, valgus, forward and backward movement.The system includes rocker system, fixed system, reward system, PC (personal computer) system, subordinate computer microcontroller system, video surveillance system, etc. After task decomposition training, monkeys can learn to use the wrist to complete the rocker behavior task quickly. The system successfully achieves the collection and recording of the wrist movement signals of monkeys.
Secondly, some key techniques of recording cerebral cortical neurons in rhesus monkeys were explored in this paper. Two kinds of mechanical devices were designed to fix the head of monkeys. The advantages and disadvantages of the two kinds of interface Utah electrodes are compared and analyzed. The fixed base for ICS-96 interface Utah electrodes is designed and improved, and the precise implantation of Utah electrodes in M1 region is realized. Based on the microelectrode array implantation technique, high quality nerves can be obtained from the motor cortex of monkeys for a long time. Cluster spike signal (spike).
Thirdly, the spike signals in M1 region and the wrist motion signals were collected synchronously, and the correlation between them was analyzed qualitatively. Firstly, the relationship between the firing pattern of nerve clusters and the direction of wrist movement was explored. It was found that the spike firing rate of nerve clusters changed most before and after the beginning of rocker movement. Secondly, the relationship between neural signals and locus of motion is analyzed by neural visualization algorithm, and it is found that neural signals have strong inherent regularity. The neural locus obtained after dimension reduction has obvious distinguishability. It can reflect the actual movement law very well.
Finally, the accurate prediction of the direction, position and speed of the monkey's wrist movement is realized by decoding algorithm with the spike firing rate of neurons as input feature. In addition, various factors affecting the decoding effect are analyzed systematically. In this paper, K-nearest neighbor algorithm (KNN) and support vector machine algorithm (SVM) are selected respectively. Vector machine (SVM) can decode and predict the movement direction of monkey's wrist, and the prediction accuracy can reach 96%. In addition, Kalman Filter (KF) and General Regression Neural Network (GRNN) are used to decode and analyze the motion parameters such as position and speed. The maximum decoding correlation coefficients (CC) of the position and speed in X and Y directions can reach 0.9170 (+ 0.0458), 0.8872 (+ 0.0778), 0.8254 (+ 0.0798) and 0.8376 (+ 0.0915).
In summary, the wrist motion parameters can be well predicted by using the neural signals recorded in the M1 region of the monkey. The wrist decoding system based on the M1 region of the monkey is a successful BCI system. The neurobiological mechanism of action laid the foundation.
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2012
【分类号】：TN911.7;R318.04
本文编号：2192462
[Abstract]:Brain-computer interface (BCI) refers to the establishment of a new type of communication and control channel between the brain and external devices, which is independent of the conventional spinal cord or peripheral neuromuscular system, so as to achieve direct interaction between the brain and the outside world. The key problem is how to obtain high-quality neural and motor signals for a long time, and how to decode and predict them by constructing decoding algorithms. Some key techniques such as motion signal recording have been explored and an implantable brain-computer interface system based on rhesus monkey wrist movement has been constructed.
Firstly, this paper designs a wrist motion signal acquisition system, which is used to collect the movement parameters of monkey wrist during varus, valgus, forward and backward movement.The system includes rocker system, fixed system, reward system, PC (personal computer) system, subordinate computer microcontroller system, video surveillance system, etc. After task decomposition training, monkeys can learn to use the wrist to complete the rocker behavior task quickly. The system successfully achieves the collection and recording of the wrist movement signals of monkeys.
Secondly, some key techniques of recording cerebral cortical neurons in rhesus monkeys were explored in this paper. Two kinds of mechanical devices were designed to fix the head of monkeys. The advantages and disadvantages of the two kinds of interface Utah electrodes are compared and analyzed. The fixed base for ICS-96 interface Utah electrodes is designed and improved, and the precise implantation of Utah electrodes in M1 region is realized. Based on the microelectrode array implantation technique, high quality nerves can be obtained from the motor cortex of monkeys for a long time. Cluster spike signal (spike).
Thirdly, the spike signals in M1 region and the wrist motion signals were collected synchronously, and the correlation between them was analyzed qualitatively. Firstly, the relationship between the firing pattern of nerve clusters and the direction of wrist movement was explored. It was found that the spike firing rate of nerve clusters changed most before and after the beginning of rocker movement. Secondly, the relationship between neural signals and locus of motion is analyzed by neural visualization algorithm, and it is found that neural signals have strong inherent regularity. The neural locus obtained after dimension reduction has obvious distinguishability. It can reflect the actual movement law very well.
Finally, the accurate prediction of the direction, position and speed of the monkey's wrist movement is realized by decoding algorithm with the spike firing rate of neurons as input feature. In addition, various factors affecting the decoding effect are analyzed systematically. In this paper, K-nearest neighbor algorithm (KNN) and support vector machine algorithm (SVM) are selected respectively. Vector machine (SVM) can decode and predict the movement direction of monkey's wrist, and the prediction accuracy can reach 96%. In addition, Kalman Filter (KF) and General Regression Neural Network (GRNN) are used to decode and analyze the motion parameters such as position and speed. The maximum decoding correlation coefficients (CC) of the position and speed in X and Y directions can reach 0.9170 (+ 0.0458), 0.8872 (+ 0.0778), 0.8254 (+ 0.0798) and 0.8376 (+ 0.0915).
In summary, the wrist motion parameters can be well predicted by using the neural signals recorded in the M1 region of the monkey. The wrist decoding system based on the M1 region of the monkey is a successful BCI system. The neurobiological mechanism of action laid the foundation.
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2012
【分类号】：TN911.7;R318.04
【引证文献】
相关博士学位论文 前1条
1 廖玉玺;植入式脑机接口神经元锋电位的时变特征分析与解码研究[D];浙江大学;2014年
，本文编号：2192462
本文链接：https://www.wllwen.com/yixuelunwen/swyx/2192462.html
