| The motor function restoration of paralyzed limbs is a challenging task. Using an electronic system as the substitute of a segment of damaged nervous tissue to rebuild the motor neuron pathways for the motor function restoration may be considered as an alternative way for this task. This method has been named as the "Micro-Electronic Neural Bridge" (MENB).Based on our research group’s prior study, this paper has focused on the study of the MENB, aiming for motor function restoration of paraplegia patients caused by spinal cord injury, and the "Micro-Electronic Muscular Bridge" (MEMB), which is aiming for the motor function restoration of the hemiplegia patients caused by stroke. Specific work has been presented as follows:1) The study of the neuronal extracellular recording simulation data generation algorithm, which is the fundation of the related neural signal processing technique.2) The study of the neural spike detection algorithm and the spike sorting algorithm, which is the fundation for the selective stimulation of the MENB.3) The prototype system design of the MENB, including the neural signal detecting circuit, functional electrical stimulation circuit, wired/wireless neural signal recording system, functional electrical stimulating signal generation system, small sized MENB experimental box, rapid prototyping algorithm verification platform, and real-time neural spike detecting and sorting system.4) The prototype system design of the MEMB, including the 2-channel MEMB prototype system which has already been used in clinical trials, and the hand motor function restoration system for hemiplegia patients caused by stroke.This paper is divided into 7 chapters. Chapter 1 discusses the background of the paralysis, introduces the background of the spinal cord injury and the stroke, and summarizes the work of the prior graduated PhD students in the research group. Chapter 2 introduces the neurophysiological fundament of the MENB, and the biological methods for the function restoration after the spinal cord injury. The principle of the MENB has also been illustrated. Chapter 3 introduces the neural signal detecting method and stimulation method of the MENB. The system design and signal processing flow are also included.Chapter 4 concerns the core signal processing technique of the MENB:the spike detection and sorting algrorithm. In this chapter, the neuronal extracellular recording simulation data generation algorithm has been discussed at first. In the study of the spike detection algorithm, the reason leading to a high false-positive rate of the traditional amplitude threshold method will be analyzed and a novel detection method based on amplitude threshold and dynamic first order forward difference threshold will be presented. With the time-domain feature filter, the average sensitivity and specificity of the algorithm has reached to 99.27% and 98.60%, respectively. As to the spike sorting algorithm design, three features extraction methods will be considered:time-domain feature, K-L domain feature, and the discrete wavelet domain feature. Two methods will be used for the feature reduction after discrete wavelet transformation. The first method is based on the Lilliefores test for normality, and the second is based on the Parzen window probability estimation. According to the separable measures in this paper, features based on discrete wavelet transformation will be used as the input of the cluster. With the K-means clustering method based on the Mahalanobis distance, the average classification accuracy can reaches to 99.29%. The self-organizing mapping will be used as a method for determining the cluster number and the initial condition.Chapter 5 mainly concerns the prototype system design of the MENB, including the neural signal detecting circuit, the functional electrical stimulation circuit, the wired/wireless neural signal recording system, the functional electrical stimulating signal generation system, and the small sized MENB experimental box. The design of the rapid prototyping algorithm verification platform based on the ARM CortexA8+TMS320C64x+MSP430F5336 hardware architecture will be discussed in details. Based on this platform, the algorithm for the real-time MENB spike detecting and sorting in bridging phase has been verified. This algorithm will be tested with 3-fold cross-validation, and the result shows that the average sensitivity, the specificity, and the identification accuracy can reach to 99.43%,97.13%, and 92.58%, respectively. Finally, the verified algorithm will be implemented on the target security MCU.Chapter 6 preliminarily discusses the MEMB, which is used for the motor function restoration of hemiplegia patients caused by stroke. This chapter introduces the principle of the MEMB, and the prototype system design, including the 2-channel MEMB prototype system, and the prototype system for hand motor function restoration. The design of myoelectricity signal detecting circuit, high-voltage, isolated, and functional electrical stimulation circuit, related signal processing algorithm, the virtual reality system for rehabilitation, and the safety consideration for clinical usage, will be introduced.Finally in Chapter 7, the work of this paper has been summarized and further problems need to be studied has also been listed.Specific innovation points of this paper are listed as follows:1) A novel spike detection algorithm based on amplitude threshold and dynamic first-order forward difference threshold has been presented. Tested by the database provided by Caltech, the average sensitivity and specificity reach to 99.45% and 97.21%, reapectively.2) A novel background noise estimation algorithm based on sliding window, range threshold, and median estimation has been proposed. Compared to the estimation method proposed by Donoho and Johnstone, the method in this paper improves the estimation accuracy three times in the situation of 100 Hz firing rate.3) Two methods for feature selection and reduction after discrete wavelet transformation have been proposed. The first method is based on the Lilliefores test for normality, and the second is based on the Parzen window probability estimation. According to the separable measures in this paper, these feature selection and reduction methods are superior to the time-domain and K-L transformation method.4) The self-organizing mapping has been used as a method for determining the cluster number and the initial condition in K-means clustering.5) Based on the board support package after Matlab R2013a, a rapid prototyping algorithm verification platform with Beagleboard-xm and Simulink as the core has been demonstrated. Real-time spike detection and sorting algorithm have been verified with this platform, and finally implemented on the target hardware. This rapid prototyping algorithm verification platform and design method can also be widely used in other signal processing filed for the algorithm design and verification. |
PDF Full Text Request