Training neural networks for stimulus artifact reduction in somatosensory evoked potential measurements

Somatosensory evoked potentials (SEPs) are useful in evaluating the integrity and determining the physiological parameters of the nervous system. Several methods have been proposed to reduce the interference in SEP measurements. Synchronous averaging is useful to reduce random interference, and adaptive methods have been developed to reduce the stimulus artifact. Since artifact and SEP often overlap in time, it has been found that it is useful to train the adaptive filter on the section of artifact before the onset of SEP. An adaptive noise canceller using a feedforward neural network is used to reduce stimulus artifact, and performance is evaluated when training is carried out on only the initial samples of artifact