005  Automated classification of inter-ictal EEG in the diagnosis of epilepsy

Background Diagnosis of epilepsy is based on descriptions of attacks, supported by finding epileptiform activity on EEG. Routine EEG can detect inter-ictal epileptiform discharges (IEDs) in approximately 50% of patients with epilepsy, and approximately 1% of the general population can have IEDs without ever developing epilepsy. The lack of a reliable biomarker is a factor in the high misdiagnosis rate of epilepsy. The EEG carries a wealth of information on brain activity, which cannot be discerned by visual analysis. Advanced EEG analysis with machine learning approaches hold promise in unlocking additional useful information contained in the EEG signal. Methods 20 patients with idiopathic generalised epilepsy and 20 age-matched controls underwent a 64-channel EEG. After pre-processing, a 20 second epoch of resting state, eyes closed EEG free from IEDs and artefact was used for analysis. Band-pass filtered EEG was used to construct functional connectivity matrices by calculating phase locking value (PLV) between electrode pairs. Connectivity matrices were analysed by a machine learning algorithm to classify patients into epilepsy and controls. Results The ML classifier achieved a classification accuracy of 60% between patients with epilepsy and normal controls. Conclusions Automated machine learning analysis has the potential to improve the diagnostic utility of EEG in epilepsy. Alternatives to functional connectivity matrices will be explored in future studies with the objective of improving classifier performance. rohanahmed92@gmail.com