[Improving signal-noise relations of evoked potentials by neural networks]

Single trial evoked potentials recorded in a sleep deprivation experiment were analyzed by means of neural networks. For improving the signal-noise-ratio a neural network classifier was trained to distinguish between the S-class which consists of assumed visually evoked single trial potentials and the N-class which should contain no event related components. The members of S-class and the N-class were averaged separately. We compared the averaged potentials of all single trials with the averaged potentials of the S-class members and we were able to show a significant improvement in the amplitudes and the signal-noise-ratio. We got the result that increasing sleep deprivation corresponds to a significantly decreasing number of members in the S-class. We concluded that sleep deprivation led to a significant number of stimuli which could not be preprocessed.