Autoregressive features based classification for seizure detection using neural network in scalp Electroencephalogram

The seizure detection in Electroencephalogram (EEG) of epileptic patients is complex as the events can be rare and unpredictable. This paper describes an autoregressive model for the identification of epileptic seizures in EEG using Artificial Neural Networks (ANN). The architecture is a three-layered feed forward network with six input nodes, two output nodes for the two classes – normal and seizure. The trained network gave a sensitivity and specificity of 91/ and 96/ respectively on test data for seizure detection. The autoregressive model therefore suggests a potential technique for pre-processing of EEG signals to be subsequently used as features for classification.