摘要
神经元锋电位可靠检测在神经科学研究与脑机接口应用中具有重要价值.针对低信噪比条件下锋电位检测的假阳性问题,提出了一种利用锋电位信号背景噪声能量和服从卡方分布约束的虚假锋电位去除方法.首先使用K-Means算法对过阈值检测的待判锋电位进行聚类,并用最小协方差算法估计各聚类总体噪声均值向量与协方差;进而计算各噪声样本与对应总体之间的马氏距离平方作为锋电位背景噪声能量和的度量指标;最后利用该指标卡方分布的置信区间对虚假锋电位进行筛除.不同信噪比条件下的仿真数据和动物实验数据应用结果表明:与传统的基于锋电位波形特征的阵列去噪算法相比,该方法可以有效识别出单电极记录神经信号中的虚假锋电位,正确率在95%以上,并且计算结果不依赖于聚类参数的选择.
The reliable detection of neuronal spikes plays an important role from basic research in neuroscience to brain-machine interface applications. In order to solve the false positive problem in spike detection, a method was proposed to remove fake spikes, by using the chi-square distribution constraint of noise energy sum. First, the detected spikes over the threshold were separated by K-Means clustering, and the noise sam- pies were acquired so that its means and covariance could be estimated by minimum covariance determinant （MCD） algorithm. Then, the Square of Mahalanobis Distribution （SMD） between each noise event and corre- sponding population was calculated to indicate the energy sum of noise. Finally, fake spikes were identified if their SMD value were not included in the confidence interval of corresponding chi-square distribution. The resuits from synthetic data and real neural data showed that the de-noising performance of this method is superior to the traditional methods. Its accuracy rate to identify the fake spikes is above 95% , and its performance is not dependent on the choice of clustering number.
出处
《郑州大学学报（工学版）》
CAS
北大核心
2015年第5期111-115,共5页
Journal of Zhengzhou University（Engineering Science）
基金
国家自然科学基金资助项目(U1304602
61473266
61305080)
河南省高等学校重点科研资助项目(15A120016)
