Neural networks for classification of multichannel EEG signals
D.C. Reddy, K. Deergha Rao · Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society · 1992
A neural network approach for the classification of multichannel EEG signals is presented. The EEG signal is split into segments and transients by means of a technique known as Linear Prediction. The linear predictor coefficients that vary from segment to segment are considered as features of a segment and a set of grapho elements as features of a transient. These features are fed as input to a 3 layered neural network for classification of EEG signal. Training of neural netork is achieved through the back propagation algorithm. EEG data from one of the channels was used to train the network. Since EEG data of one channel invariably differs from the other over the same interval of observation, one channel data was used for training while the other channels data for testing. This also enables one to have inter- and intra-channel comparisons. The network was then tested for K-complexes and sleep spindles. It was found that the network was able to successfully recognise these patters. Testing of the network with transients, background noise and unspecified activity are currently under investigation.