Estimation For Evoked Potential Signal With Normalized Radial Basis Function Neural Network

The study of different sensitive or cognitive functions, states, pathologies, etc, makes the evoked potentials (EP) an invaluable tool in neurophysiology. Of particular interest was extracting information about the potentials from measurements that contain also on-going background electroencephalogram (EEG). However, as a traditional application of signal estimation, no approaches can achieve desired estimates for single-trial potential yet. In the present paper, we consider normalized radial basis function neural network (NRBFNN). This scheme is achieved to out-perform other traditional methods in the case of determining the temporal relationship of variance between trials and measuring response synchronously to each stimulus.Firstly we review the universal backgrounds of current technologies on the single-trial estimation, and introduce some of these cases. Compared with other methods, NRBFNN is more specialized in the ability of unbiased estimation in the case of nonlinear signals. It can filter any in-correlate noise overlapped on the EP signals and finally yield smooth interpolation curve represented as desire signals. Compared some of techniques proposed, simulations results illustrate that our approach can attenuate noise with 1-2 trials input, and give variance between trials at the same time. Trail-to-trial variance is meaningful to clinicians and also meaningful in recognition problems. NRBFNN is also conducted to investigate ERPs' correlation of the scale to visual spatial attention. A significant main effect of spatial cues reflecting on P300 is identified, which is neglected by former researchers. It can be predicted that NRBFNN will make more achievement in the future.