A time domain frequency-selective multivariate Granger causality approach.
A time domain frequency-selective multivariate Granger causality approach.
复制标题
时域频率选择性多元格兰杰因果关系方法
DOI：
10.1109/embc.2016.7591968
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
Leistritz
中科院分区：
文献类型：
--
作者：
Leistritz
有效连接性的研究是计算神经科学中的一个重要课题，它可以用来理解大脑中空间分布的神经元之间的相互作用。因此，在过去的几十年中，已经开发了各种各样的方法来研究多变量系统中的功能和有效连接。它们的频谱范围从基于模型的方法到无模型方法，并明确分为时间和频率范围方法。我们在这个模拟研究中提出了一种新的时域方法的基础上格兰杰的可预测性原则，它允许频率选择性的考虑定向相互作用。它是基于多变量自回归模型拟合系统修改的时间序列的预测误差的比较。这些修改是基于信号分解，这使得能够有针对性地消除具有特定频谱特性的特定信号分量。根据嵌入信号分解方法，可以导出频率选择性或数据驱动的信号自适应格兰杰因果指数。
The investigation of effective connectivity is one of the major topics in computational neuroscience to understand the interaction between spatially distributed neuronal units of the brain. Thus, a wide variety of methods has been developed during the last decades to investigate functional and effective connectivity in multivariate systems. Their spectrum ranges from model-based to model-free approaches with a clear separation into time and frequency range methods. We present in this simulation study a novel time domain approach based on Granger's principle of predictability, which allows frequency-selective considerations of directed interactions. It is based on a comparison of prediction errors of multivariate autoregressive models fitted to systematically modified time series. These modifications are based on signal decompositions, which enable a targeted cancellation of specific signal components with specific spectral properties. Depending on the embedded signal decomposition method, a frequency-selective or data-driven signal-adaptive Granger Causality Index may be derived.