Thorbergsson P T, Jorntell H, Bengtsson F, Garwicz M, Schouenborg J, Johansson A
Neuronano Research Center, Dept. of Electrical and Information Technology, Lund University, Lund, Sweden.
Annu Int Conf IEEE Eng Med Biol Soc. 2009;2009:6998-7001. doi: 10.1109/IEMBS.2009.5333847.
A well defined set of design criteria is of great importance in the process of designing brain machine interfaces (BMI) based on extracellular recordings with chronically implanted micro-electrode arrays in the central nervous system (CNS). In order to compare algorithms and evaluate their performance under various circumstances, ground truth about their input needs to be present. Obtaining ground truth from real data would require optimal algorithms to be used, given that those exist. This is not possible since it relies on the very algorithms that are to be evaluated. Using realistic models of the recording situation facilitates the simulation of extracellular recordings. The simulation gives access to a priori known signal characteristics such as spike times and identities. In this paper, we describe a simulator based on a library of spikes obtained from recordings in the cat cerebellum and observed statistics of neuronal behavior during spontaneous activity. The simulator has proved to be useful in the task of generating extracellular recordings with realistic background noise and known ground truth to use in the evaluation of algorithms for spike detection and sorting.
在基于中枢神经系统（CNS）中长期植入的微电极阵列进行细胞外记录来设计脑机接口（BMI）的过程中，一套定义明确的设计标准非常重要。为了比较算法并评估它们在各种情况下的性能，需要有关于其输入的真实情况。如果存在最优算法，从真实数据中获取真实情况就需要使用这些算法。但这是不可能的，因为这依赖于有待评估的算法本身。使用记录情况的现实模型有助于细胞外记录的模拟。该模拟能够获取先验已知的信号特征，如尖峰时间和身份。在本文中，我们描述了一种模拟器，它基于从猫小脑记录中获得的尖峰库以及自发活动期间神经元行为的观测统计数据。事实证明，该模拟器在生成具有现实背景噪声和已知真实情况的细胞外记录以用于评估尖峰检测和分类算法的任务中很有用。
