脑电ERP信号的特征提取与应用.PDFVIP

脑电/ERP信号的特征提取与应用 ——ERP研究与应用研讨会 胡 理 西南大学副教授 Website: Email: hulitju@ Event-related EEG potentials (ERPs) – Basics of averaging Across-trial averaging and the additive noise model Event-related EEG potentials (EPs) – Multichannel recording When more than one electrode is used, the average EPs can be plotted according to the relative position of each electrode on the head, thus providing some spatial information on potential distribution. Event-related EEG potentials (EPs) – Overview 200-250 ms 350-400 ms 500-600 ms Peaks are characterised by their latency, polarity, amplitude and scalp topography Across-trial averaging 传统的ERP分析方法能有效提高脑响应信号的信噪比，但该方法导致了 有用信息的大量丢失: 单个被试：几十/几百M （原始数据） 几十/几百K （叠加波形） 几十/几百bit （提取用于统计的参数） 1、损失了什么信息？ 2、用什么办法可以挖掘更多信息？ 1. 动态信息——脑电信号单次分析（single-trial analysis ） 2. 节律信息——脑电信号时频分析（time-frequency analysis ） 3. 网络信息——脑电信号连通性分析（connectivity analysis ） ERP动态信息 Problem #1: Variability in latency Problem #2: Variability in amplitude Problem #3: Variability in morphology Variability of single-trial latency, amplitude… and morphology 脑电信号单次分析方法 Time domain Approach #1: Spatial filtering: Probabilistic ICA (Hu et al Clinical Neurophysiology 2011) Common spatial pattern (Huang, Xiao, Zhang, Hu. NeuroImage 2013) Approach #2: Wavelet filtering (Hu et al NeuroImage 2010) Approach #3: Parametric approach (Hu et al Journal of Neurophysiology 2011) Toolbox: STEP1 () No detection bias (assessed on a real ERP dataset) Correlation with behavioural measures 被试内ERP与行为学的关联分析 Single-trial EEG-fMRI 的融合 ……