Neural network classification of EEG data generated through use of the audio oddball paradigm

Single trial electrical brain evoked potentials were recorded using two different audio stimuli with unequal probabilities of occurrence. A modified self-organizing Kohonen neural network was used to organize the responses. The output of the network formed a topological map of the various component responses which included both changes in the evoked potential waveforms and the performance reaction times. The classification rate was over 88% which compares favorably with the expert human. The algorithm serves as an automatic intelligent classifier for electroencephalograph (EEG) signals and could be applied as a powerful research tool in cognitive mode mapping or workload analysis