Automated Detection of Epilepsy using a LAMSTAR Neural Network

V.P. Nigam and D. Graupe (USA)


Epilepsy, medical diagnosis, Neural Networks (NN),LAMSTAR NN


The benefits of performing a first pass on the data with an automatic seizure detector are clear -- If the amount of data presented to the technician can be reduced then more patients can be treated. Even within the in-patient recording environment an automatic detection system can prove advantageous. Salinsky has reported, in a study (1997) of 83 patients admitted for in-patient monitoring, that 22% of the seizures recorded were captured only by the computer detection system, resulting in an estimated saving of 1.3 hospital days per admission (see [1]). The diagnosis of epilepsy is primarily based on scalp recorded electroencephalograms (EEG). Unfortunately the long-term recordings obtained from "ambulatory recording systems" contain EEG data of up to one week duration, which has introduced new problems for clinical analysis. Traditional methods where the entire EEG is reviewed by a trained professional, are very time consuming when applied to recordings of this length. This paper describes detection of epileptic seizures from EEG signals using a multistage nonlinear preprocessing filter and a diagnostic (LAMSTAR) Artificial Neural Network (ANN) to reduce expert effort in analyzing of lengthy recordings. Preprocessing via multistage nonlinear filtering, LAMSTAR input preparation, ANN training and system performance (1.6% miss rate, 97.2% total accuracy) are discussed. In case of epilepsy, two categories of abnormal activity are observed in the EEG: Ictal (during an epileptic seizure) and inter-ictal (between seizures) [1]. The most common form of inter-ictal abnormality is spiking (in form of individual spikes, spike trains or spike and wave features). These spikes are seen in the majority of epileptic patients, whereas only a very small number of non-epileptic patients show this feature. For this reason inter-ictal spike detection plays a major role in the diagnosis of epilepsy. However, during an isolated spike, the brain is not in a clinical seizure. A very different EEG pattern is seen during the ictal period consisting of rhythmical waveforms. Although inter-ictal spikes offer evidence of epilepsy this can only be confirmed and a complete diagnosis made, by an observed seizure. Hence detection of epilepsy boils down to the problem of recognizing EEG spikes and classifying them as epileptic or normal.

Important Links:

Go Back