Recognition of Epileptic Activity on the Basis of EEG Using Support Vector Machines
Andrzej Rysz · 2004
The electroencephalogram (EEG) of the person who suffers from epilepsy is characterized by the occasional spikes between seizures. They are not easy to detect even for clinicians. Therefore the automatic computerized methods are needed. The paper presents the solution to this problem by applying Support Vector Machine (SVM) network. The important stage of this approach includes the generation of the features on the basis of which SVM will recognize all spikes appearing in the registered EEG. The results of numerical experiments will be presented and discussed in the paper I. INTRODUCTION The epilepsy is defined as a chronic brain disorder of various aetiologies characterized by recurrent seizures (ictal disturbances) due to the excessive discharge of cerebral neurons (2,3). Between seizures, the EEG of subjects who suffers from epilepsy is characterized by the occasional inter-ictal activity in the form of so called spikes and wave complexes. A spike is a sudden burst of electrical activity lasting up to 70 ms. The sharp wave is sort of spike of longer duration, usually between 70ms and 200ms. At the EEG recording of the suspected epileptic patient we almost always record the inter-ictal spikes. Only occasionally a seizure may occur directly at the hospital investigation and be recorded by EEG. Hence epilepsy is confirmed mainly on the basis of inter-ictal recordings found in EEG. Unfortunately the inter-ictal activity is usually relatively rare and may be missed by the clinicians, who check the recording of the suspected patient at the routine inspection. Therefore the need for an automated analysis systems, which can reliably identify spikes in the recorded EEG waveform and recognize between spiky and non-spiky types, are highly required. This paper will present the solution of an automatic system for detecting inter-spike activity by applying the time and frequency analysis of the recorded EEG waveforms and using the Support Vector Machine (SVM) as the final recognizing system. The recognition of epileptic activity by SVM will use the features generated on the basis of EEG waveform analysis. The SVM network itself performs the separation of the data corresponding to spiky and non-