Barlow J S
Department of Neurology, Massachusetts General Hospital, Boston 02114.
J Clin Neurophysiol. 1985 Jul;2(3):267-304. doi: 10.1097/00004691-198507000-00005.
Methods for analysis of nonstationary EEGs, that is, EEGs whose patterns undergo changes with time (e.g., alpha blocking, paroxysmal slow waves, onset of drowsiness/sleep, but excluding spikes/sharp waves) are reviewed. The concepts of stationarity and nonstationarity, and general techniques for their evaluation, are discussed. Simpler methods for monitoring for nonstationarity include running determinations of average amplitude and average period or interval. Piecewise stationary analysis includes characterization, by spectra obtained by fast Fourier transform or by autoregressive modeling, of sections of EEGs preselected to be stationary. In Kalman filtering, the autoregressive model itself becomes time-varying. Segmentation of the EEG into stationary lengths can be carried out on a fixed-interval basis (i.e., of successive, e.g., 1-s intervals), with clustering (grouping) or classification according to the features (e.g., spectra) of each interval, and concatenation of adjacent similar intervals. Alternatively, in adaptive (variable-interval) segmentation, the EEG is continuously monitored automatically for any significant departure from stationarity, and segment boundaries are placed accordingly. A number of applications of the various methods are included, with examples of succinct summary displays. Problems and prospects are discussed.
本文综述了非平稳脑电图（EEG）的分析方法，即其模式随时间变化的脑电图（例如，α波阻断、阵发性慢波、嗜睡/睡眠开始，但不包括棘波/尖波）。讨论了平稳性和非平稳性的概念及其评估的一般技术。监测非平稳性的较简单方法包括对平均幅度和平均周期或间隔进行连续测定。分段平稳分析包括通过快速傅里叶变换或自回归建模获得的频谱对预先选择为平稳的脑电图部分进行特征描述。在卡尔曼滤波中，自回归模型本身随时间变化。脑电图可按固定间隔（即连续的，例如1秒间隔）划分为平稳段，根据每个间隔的特征（例如频谱）进行聚类（分组）或分类，并将相邻的相似间隔连接起来。或者，在自适应（可变间隔）分段中，对脑电图进行连续自动监测，以检测任何与平稳性的显著偏差，并相应地设置分段边界。文中包含了各种方法的一些应用，并给出了简洁汇总显示的示例。还讨论了存在的问题和前景。
