基于关联维数的神经元动作电位特征提取与分类研究  

Study on Feature Extraction and Spike Sorting Based on Correlation Dimension

作　　者：詹跃荣[1] 范影乐[1] 杨文伟[2] 杨勇[2] 李轶[2] 

机构地区：[1]杭州电子科技大学自动化学院,杭州310018 [2]杭州电子科技大学生物医学工程与仪器研究所,杭州310018

出　　处：《中国生物医学工程学报》2011年第5期648-654,共7页Chinese Journal of Biomedical Engineering

基　　金：浙江省新苗人才计划项目(2009G60G2040018)

摘　　要：神经元动作电位模式分类是动作电位序列分析和解码的基础。由于神经元动作电位信号波形隐藏着动作电位的特征信息,而关联维数是度量波形不规则程度的一种手段,因此基于神经元动作电位信号的波形差异,提出一种基于关联维数对神经元动作电位进行特征提取的新方法。首先,对采集到的神经元动作电位信号进行相空间重构;然后,在重构的相空间中,以关联维数作为对非同源动作电位信号的特征进行描述;最终,结合K均值算法,对神经元动作电位实现无监督模式分类。仿真和真实数据实验结果表明:该方法分类的准确率较高,且稳定性较好,仿真数据分类错分率基本稳定在4.9%以内,真实数据的分类能较大程度地贴近人工分类的结果,因此用来代替人工分类具有一定的可行性。Neural spike sorting is the basis of spike train analysis and decoding.With neural spike waveform hiding its feature information,and correlation dimension as a way of measuring the extent of irregular wave,we proposed a new method of feature extraction to spike based on correlation dimension.At first,we reconstructed phase space for collected neural spike.Then,the correlation dimension was used as the feature description of non-homologous spikes in the reconstructed phase space.Finally,unsupervised spike sorting was achieved by using K-means algorithm.The results demonstrated that the spike sorting method achieved higher accuracy and better reliability.Experiments on simulated spike signals showed encouraging misclassified rate below 4.9%.Furthermore,experiments on real spike signals showed that the clustering results highly agreed with manual sorting.Therefore the sorting method has feasibility of replacing manual sorting.

关 键 词：动作电位分类 关联维数 相空间重构 K均值 

分 类 号：R318[医药卫生—生物医学工程]