AUTOMATIC DETECTION OF EPILEPTIFORM SPIKES IN EEG BASED ON TIME-FREQUENCY ANALYSIS
基于时频分析检测EEG中癫痫样棘/尖波的方法

A new method for detecting spikes in the EEG based on the time-frequency representation is presented. The EEG records are divided into epochs having duration of 5s, and then the time-frequency representation of a segment of the EEG signals is evaluated by using Choi-Williams distribution in order to be able to overcome the drawbacks of the cross-terms. We define the spectral components at any time instant in the time-frequency distribution of a signal as the signal's instantaneous spectrum. The mean values of all the instantaneous spectra in a segment of EEG signal are used to describe the background activity of the EEG over the epoch. The quantitative difference between the instantaneous spectrum at each instant and the mean instantaneous spectrum is called the spectral difference. Based on the experiments, we design a nonlinear transform for processing the spectral difference. Spikes are detected by the results of the transform of spectral difference. The results show that the new method works well.