Detection and classification of neural information from peripheral nerves

功能性电刺激（FES）用于恢复脊髓损伤或脑卒中引起的自主运动功能丧失。目前的功能性电刺激系统存在的问题是，由于开环控制系统，在存在干扰和肌肉疲劳的情况下，恢复效果较差。为了开发具有天然传感器的闭环FES系统，我们研究了生物相容性和低噪声电极，以及从周围神经中提取神经信息的方法，研究了不锈钢SUS316L、钴铬基合金NAS604PH和新开发的高氮高锰奥氏体不锈钢NAS106N的疲劳性能和噪声特性。结果表明，在空气和生理盐水中，NAS106N电极的疲劳寿命均高于SUS316L和NAS604PH电极。我们发现郎维结和记录电极之间的距离影响神经动作电位的记录波形，并且即使每个纤维的直径和膜的性质相似并且每个纤维在神经干中接近，也可以基于记录的波形来识别每个神经纤维的活动。自然传感器闭环FES系统在临床应用中，记录装置必须放置稳定，对神经纤维损伤小。袖带电极满足这些要求。然而，它有时会提供信噪比太低而无法检测动作电位的记录。我们采用小波变换在低信噪比条件下检测动作电位。我们的方法提供了可靠的结果。此外，我们还提出了用聚类分析来估计单个单位活动的动作电位分类方法，以及确定动作电位分类数目的标准。它提供了可靠的结果。

The functional electrical stimulation (FES) is used to restore the loss of voluntary motor function caused by the spinal cord injury or the cerebral apoplexy. Less effective for restoration in the presence of the disturbance and muscle fatigue because of open-loop control system are problems of the current FES system. For developing closedloop FES system with natural sensors, we studied biocompatible and low noise electrodes, and method extraction of neural information from peripheral nerves.Fatigue properties and noise characteristic of stainless steel SUS316L, Co-Cr based alloy NAS604PH and newly developed high nitrogen high manganese austenitic stainless steel NAS 106N were investigated. We found that each electrode did not generate excess noise, and NAS106N had higher fatigue life than SUS316L and NAS604PH in both air and saline solution.A model of peripheral nerve was developed. We found that the distance between Ranvier's node and the recording electrode influenced on the recorded waveforms of nerve action potentials and it was possible to identify activity of each nerve fiber based on the recorded waveforms even if diameter and property of membrane of each fiber were similar and each fiber was close in nerve trunk. For clinical application of closed-loop FES system with natural sensors, the recording device must be placed stably and give minor damage to nerve fiber. Cuff electrode satisfies these requirements. However, it sometimes provides recordings whose signal-to-noise ratio is too low to detect action potentials. We used wavelet transform to detect action potentials under low signal-to-noise ratio condition. Our method provided reliable result. Moreover, we proposed the method for classification of action potentials to estimate single-unit activity with cluster analysis and the criterion for deciding the number of groups that action potentials were classified into. It provided reliable result.