以改良式單一連結法進行非監督式神經動作電位辨識分類

Abstract

神經動作電位辨識分類是許多神經科學研究的基礎之一，其目的是以神經元胞外動作電位訊號做為研究對象，並將不同神經元產生的神經動作電位做分門歸類。在記錄神經元訊號時，通常會記錄多筆且冗長的神經元訊號，亦會有同時記錄到多個神經元放電的情形，在大量的神經元訊號處理下，神經動作電位辨識分類會變的緩慢且沈重，因此神經動作電位辨識分類通常需要大量的人工操作來完成。而大多數的研究都只注重分類方法的精確度和在模擬訊號下的效果，前者會導致在高準確率的情形下缺乏分類方法的便利性，而後者更容易缺少實際應用情形的測試來輔助，如此一來會使神經動作電位辨識分類方法的使用上造成不便。 本研究針對多神經元記錄訊號，使用一套新的神經動作電位辨識分類方法，此方法分別以小波轉換和主成份分析法做為神經動作電位波形上特徵的截取，再以改良式單一連結法結合灰色理論做自動化的神經動作電位分類，過程均是以非監督式且合理的方法進行，來達到有效、快速且方便的神經動作電位辨識分類。此方法以良好的模擬訊號做測試，並和一些其它的分類方法比較，最後則進一步以從人體身上量測的神經元訊號做分類檢測，其結果也證實本文的方法能有效的以非監督式過程來進行神經動作電位辨識分類。

Spike sorting is the one of the based in many neuroscience researches. The targets of spike sorting are to research the action potential of neuron and to classify the difference spikes produced by difference neurons. The recorded neuron signals is often diffusion and astronomical. It may also record the signals produced by two or more neurons. A lot of neuron signals make the procedure of spike sorting slow and weighty. Hence the spike sorting needs much artificial operations to complete. Many researches attach importance to the accuracy of classification and the performance under the simulate situation. The first lead to the lack of convenience, and the later lacks the test of real applications to support. Hence the spike sorting can’t conveniently be used for signal analysis. The thesis uses a new spike sorting method for multiunit neuron signals. The method respectively use wavelet transform and principal components analysis to extract the spike waveform features. Then the spike classification method is based on improved signal linkage method and grey theory to arrive automatically procedure. The procedure is unsupervised and reasonable and hence it can make a fast, effective and convenient spike sorting method. The method is tested by strict simulation spikes and compared with some other classification methods. Subsequently the method analyzes the neuron signals recorded from human. The results also show that the method of the thesis has an unsupervised spike sorting procedure.