航天医学与医学工程

目的实现神经元峰电位(spike)的准确检测和分类,为神经信号的后续分析和解码提供前提条件。方法采用改进的阈值法,从植入式多电极阵列采集的含噪神经电信号中检测出有效的峰电位;并提出利用奇异谱熵,来描述在奇异值分解下峰电位特征;通过Kolmogorov-Smirnov检验降低特征维数,采用交互式方式挑选聚类性能较佳的二维特征向量;最后结合C均值聚类算法实现峰电位分类。结果本文提出的峰电位奇异谱熵特征使多组仿真和真实神经电生理信号获得了较为理想的聚类效果,且仿真数据分类准确率几乎都达到98%以上。结论基于奇异谱熵的峰电位特征提取,能够较好地表达和区分各类别峰电位的动态特性,可以作为峰电位有效的分类依据。

Objective To realize the accurate detection and classification of neuronal spikes,as well as provide the premise for further analyzing and decoding neural signals.Methods An improved threshold method was adopted to find out effective spikes of neuronal action potentials from the noisy electroneurographic signals,which were collected with the implanted multi-electrode array.Then based on the singular value decomposition(SVD),the features of original spikes were characterized with singular spectrum entropy(SSE).Kolmogorov-Smirnov test was used to reduce the dimensions of features.A two-dimensional feature vector was selected to gain the best result of clustering through interactive mode.Finally,the classification of neuronal spikes was achieved with C-means clustering algorithm.Results The perfect results of clustering were gained with this feature of SSE from many groups of simulated or real electroneurographic signals.The accuracy of simulated signal classification was almost above 98%.Conclusion The features based on SSE of spikes can be perfectly applied to distinguishing their dynamic characteristics and be the effective basis of classification of neuronal spikes.