Epileptic Seizure Detection Using Support Vector Machines

Negin Moghim (Heriot-Watt University and University of Edinburgh, Edinburgh)

Epilepsy is a neurological disorder which affects 50 million people worldwide, 25% of which are resistant to Anti-Epileptic drugs. Seizure detection for those patients suffering from drug-resistant epilepsy has been one of the main focuses of researchers in the field, as this could be coupled with other state of the art technology to alert patients before seizure occurs, allowing them to take the right action to prevent potential injury from seizures.

 This study aims to further investigate other ways of improving seizure detection methods and proposes that automatic seizure detection could also be used as a tool for further investigating the patterns of refractoriness among epilepsy patients.

 The study at this stage, investigates the performance of several variations of the Support Vector Machines in comparison to previous research conducted on the application of neural networks for seizure detection.

 The study is carried out on EEG recordings of two epileptic patients; two classification models are derived from each patient. The models are then tested on the same patient and the other patient, comparing the specificity, sensitivity and accuracy of each of the models.

 The features extracted from the EEG recordings, which will be used for training the SVM, are a combination of signal energy, wavelet transform and nonlinear system dynamics concept.

 The results of this stage of the study will be plugged into the bigger picture of the research: the analysis of several epilepsy patient records, in order to understand the factors that contribute to their refractoriness and providing a prognosis model for each individual patient. The next stage of this study is to compare the EEG records of patients suffering from drug-resistant epilepsy with those patients who have had successful treatments and observe whether any patterns can be found.

Preferred presentation format: Poster
Topic: General neuroinformatics

Document Actions