结合小波分解和径向基神经网络进行事件相关电位的单次提取

COMBINING WAVELET DECOMPOSITION WITH RADIAL BASIS FUNCTION NEURAL NETWORK IN ESTIMATION OF SINGLE-TRIAL EVENT-RELATED POTENTIAL

从单次实验记录中提取事件相关电位 ,无论在临床诊断上还是在大脑高级功能的研究中都起着重要的作用。介绍了一种将小波多分辨率分解和重建与径向基神经网络结合起来进行事件相关电位单次提取的方法。它基于事件相关电位主要是低频信号的事实 ,发挥径向基神经网络对连续函数的逼近能力 ,从信号的小波分解系数中提取出与低频响应相关的成分 ,构造了一种新的时频域滤波的方法 ,实验表明本方法较好地从单次记录中提取出了事件相关电位。

Single-trial estimation of ERP plays an important role both in clinic and in advanced brain function research. A method combining wavelet multi-resolution decomposition and reconstruction with radial basis function neural network (RBFNN) is presented in this paper. Based on the fact that ERP is low frequency signal and the powerful function approximation ability of RBFNN, low frequency components related to ERP were estimated by the neural network from the detail coefficients of the record, and these components were used to reconstruct the underlying ERP. With this method, we successfully estimated the underlying ERP from single-trial ERP records.