Feature extraction of event-related potential waveforms by neural networks

Artificial neural network (ANN) Methods may be useful in brain signal analysis, in which the signal characteristics are unknown and signal-to-noise ratios are well below one. The development of a neural network for classifying event-related potential data obtained from normal control subjects and from patients with multiple sclerosis is described. The classification strategy is then decoded by network analysis and compared with that obtained statistically. The network decision-making process is illustrated by three examples, showing the variation of the responses of internal hidden units to different input stimuli