Evoked potentials (EPs) are the special signals that are non-stationary and corrupted by relatively large background noise. To extract the time-varying EP responses more correctly from the noise, a new method is proposed to investigate the problem of denoising the EP signals. The main objective is to estimate the amplitude and the latency without losing the individual properties of each epoch, which is meaningful to clinicians and recognition problems. A normalized radial basis function neural network (NRBFNN) was presented to process the raw EP signals for the purpose of canceling the background noise. The output of NRBFNN enables to effectively track the EPs' variations since the proposed basis functions covers the whole input space with the same degree. Simulations and experimental results confirmed the superior performance of NRBFNN over other methods.
[1]
S Nishida,et al.
Method for single-trial recording of somatosensory evoked potentials.
,
1993,
Journal of biomedical engineering.
[2]
F. Girosi,et al.
Networks for approximation and learning
,
1990,
Proc. IEEE.
[3]
B. Widrow,et al.
Adaptive noise cancelling: Principles and applications
,
1975
.
[4]
R. Isermann,et al.
Closed-loop ignition control using online learning of locally-tuned radial basis function networks
,
1999,
Proceedings of the 1999 American Control Conference (Cat. No. 99CH36251).
[5]
James D. Keeler,et al.
Layered Neural Networks with Gaussian Hidden Units as Universal Approximations
,
1990,
Neural Computation.