使用多通道微電極陣列晶片 記錄神經元動作電位以計算其空間連結

Abstract

本論文描述了一套應用在多通道微電極陣列晶片系統上的放電神經元定位演算法及微培養系統之研發，從微培養器系統的建構、如何將偵測到的電生理訊號轉換到空間平面、並透過類神經網路分類找出放電神經元的空間位置作一個深入的探討，並建構一個模擬的神經網路來驗證演算法的改善及特性。最後以放電神經元定位演算法來討論大腦皮質神經元的連結性。 在本研究中以微培養器的MEA加熱環以及溼度控制器結合自行設計的MEA chamber適當地控制多通道微電極陣列的溫度、溼度及氣體組成，以達到可長時間觀察神經元放電活動。放電神經元定位演算法是由兩個步驟組成，第一個為先以互相關性找出單一神經元的放電訊號，並透過牛頓逼近法來求出放電神經元的近似位置，再將各電極中重複定位的結果移除掉並將剩餘的近似位置送到第二步驟，第二步驟是以類神經網路演算法找出神經元真實的放電位置。 在實驗的結果上，以模擬的神經網路放電訊號驗證有加入互相關性的放電神經元演算法的平均誤差為1.73μm，四個電極可辨識到的最大神經元數量可達六個，其透過移除最外圍電極與中心電極重複定位的方式可避免位於微電極外圍放電神經元對定位演算法之干擾，其與不考慮關聯性的定位演算法比較起來有更好的定位準確度。由於演算法本身包涵神經訊號的互相關性及放電時序的特性，本演算法可以用來探討神經元突觸的連結性。本研究的微培養器及放電神經元定位演算法可為探討長時間的神經元連結性改變之研究提供了另一種以放電神經元位置作為連結探討的選擇。

This study describes the development of spike source localization algorithm and micro-incubation system with Multi-Electrode Array (MEA). The construction of the micro-incubation system, the process of nonlinear mapping of a set of electrode amplitudes in to two-dimensional (2D) space and the process of self-organizing map clustering technique are presented in detail. In addition, the biologically realistic network simulation is used to verify the performance and characteristic of the localization algorithm. Finally, the neural connectivity of the cerebral cortex neuron is discussed by using the localization algorithm. The micro-incubator controls the temperature, humidity and gas mixture by integrating the MEA warmer, humidifier with MEA chamber for long-term observation of neuron activity. The firing neuron localization algorithm consists of two processes. The first process uses the cross-correlation to find the spike firing from the single neuron, and then the Newton method is used to compute the approximation location of the firing neuron. Next, the removal of overlapping localization between electrode groups is used to enhance localization accuracy, and the remaining approximation localizations are passed to the next process. The second process is the self-organizing map clustering technique that is used for computing the real firing neuron source location. The mean average of localization error of the localization algorithm is 1.73μm in the simulation network data, and the maximum recognition neuron number between four electrodes is six. The noise from the firing neuron outside the multi-electrodes can be avoided by using the removal process. As a result, the correlative localization algorithm performs better than the tetrode localization algorithm which is without correlation process. Since the correlation and time series between spikes are considered in the localization algorithm. The neural connectivity can be discussed by the localization algorithm. The localization algorithm research provides a tool for others research interesting in neural connectivity.