Stamoulis Catherine, Chang Bernard S
Department of Radiology, Children’s Hospital Boston and Harvard Medical School, Boston, MA 02115, USA.
Annu Int Conf IEEE Eng Med Biol Soc. 2012;2012:6157-60. doi: 10.1109/EMBC.2012.6347399.
Detection of precursory, seizure-related activity in electroencephalograms (EEG) is a clinically important and difficult problem in the field of epilepsy. Seizure detection methods often aim to identify specific features and correlations between preictal EEG signals that differentiate them from interictal/nonictal signals. Typically, these methods use information from nonictal EEGs to establish detection thresholds, and do not otherwise incorporate their characteristics into the detection. A space-time adaptive approach is proposed to improve detection of seizure-related preictal activity in scalp EEG, using multiple patient-specific baseline signals to optimize the estimate of the baseline covariance matrix. A simplified model of the preictal EEG is assumed, which describes this signal as a linear superposition of seizure-related activity and baseline activity (treated as an interference signal). It is shown that when an improved estimate of the baseline covariance is included in the preictal detector, the true positive rate increases significantly and also the false positive rate decreases significantly.
在脑电图（EEG）中检测前驱性的、与癫痫发作相关的活动是癫痫领域一个具有临床重要性且颇具难度的问题。癫痫发作检测方法通常旨在识别发作前EEG信号中的特定特征以及它们与发作间期/非发作期信号之间的相关性。通常，这些方法利用来自非发作期EEG的信息来设定检测阈值，并且在其他方面并未将其特征纳入检测之中。本文提出了一种时空自适应方法，通过使用多个患者特定的基线信号来优化基线协方差矩阵的估计，以改善头皮EEG中与癫痫发作相关的发作前活动的检测。假定了一个发作前EEG的简化模型，该模型将此信号描述为与癫痫发作相关的活动和基线活动（视为干扰信号）的线性叠加。结果表明，当在发作前检测器中纳入对基线协方差的改进估计时，真阳性率显著提高，同时假阳性率也显著降低。
