Detection and Classification of EEG Epileptiform Transients with RBF Networks using Hilbert Huang Transform-derived Features

Diagnosis of epilepsy or epileptic transients AEP (Abnormal Epileptiform Paroxysmal) is tedious, but important, and an expensive process. The process involves trained neurologists going over the patients EEG records looking for epileptiform discharge like events and classifying it as AEP (Abnormal Epileptiform Paroxysmal) or non-AEP. The objective of this research is to automate the process of detecting such events and classifying them into AEP(definitely an Epileptiform Transient) and non-AEPs (unlikely an epileptiform transient). The problem is approached in two separate steps and cascaded to validate and analyze the performance of the overall system. The first step is a detection problem to find the Epileptiform like transients (ETs) from the Electroencephalograph (EEG) of a patient. A Radial basis function-based neural network has been trained using a training set consisting of examples from both classes (ETs and non-ETs). The ETs are the yellow boxes which are marked by expert neurologists. There are no particular examples of non-ETs and any data not annotated by experts can be considered to be examples of non-ETs. The second step is classification of the detected ETs also known as yellow boxes, into AEPs or non-AEPs. A similar Radial basis function-based neural network has been trained using the ETs marked and classified into AEPs and non-AEPs manually by seven expert neurologists. The annotations or yellow boxes along with the contextual signal was used to extract features using the Hilbert Huang Transform. The system is validated by considering an entire epoch of the patient EEG and potential ETs are identified using the detector. The potential ETs marked by the detector are classified into AEPs and non-AEPs and compared against the annotations marked by the experts.