Epilepsy diagnosis from scalp EEG : a machine learning approach

Epilepsy refers to a group of chronic brain disorders characterized by recurrent unprovoked seizures. Interictal Epileptiform Discharges (IEDs) are distinctive biomarkers of epilepsy. According to the International Federation of Societies for Electroencephalography and Clinical Neurophysiology (IFSECN), IEDs are defined as transients distinguishable from background activity, with a characteristic spiky morphology, typically, but neither exclusively nor invariably, found in interictal electroencephalograms (EEGs) of people with epilepsy. Approximately 65 million people are affected worldwide. The early diagnosis of epilepsy is based on the IEDs that appear in the EEG. The EEGs are non-invasive, inexpensive, digitized recordings of brain activity from the scalp. However, there is a lack of experts who can analyze EEGs; moreover, manual interpretation EEG is time-consuming. Moreover, IEDs exhibit huge variation across patients. Besides, a standard definition of interictal epileptiform events is yet to be formulated. This makes the IED detection based diagnosis of epilepsy, tedious, and expert-centered. Experts have a significant disagreement over IED annotation. Consequently, an automated objective algorithm for detecting IEDs would be beneficial for the diagnosis of epilepsy. Our study had two primary objectives: to study the underlying characteristics of epileptiform IEDs and to develop an automated epilepsy detection system. First, we quantified the different morphological characteristics of IEDs, and a standard definition of IEDs was formulated. Later, we developed an intelligent automated epileptic EEG identification system based on interictal pattern detection. The system consists of three major parts: the preprocessing module, the interictal pattern detection module, and the EEG classification module. We have evaluated the system on a sizable routine scalp EEG database from multiple centers, namely, Massachusetts General Hospital (MGH), Medical University of South Caroline (MUSC), and National University Hospital (NUH) Singapore. The MGH dataset consists of 545 subjects: 84 epileptic patients with IEDs annotated by two neurologists and 461 non-epileptic subjects. We applied a prepossessing pipeline with the following components: data preparation, data segmentation, filtering, and artifact removal. Next we employ a convolutional neural network (CNN)-based pattern detection for identifying epileptiform patterns. The system achieved superior performance compared to conventional methods with the highest area-under-curve (AUC) of 0.989±0.006 and a false alarm rate of 0.2±0.12 per minute for a sensitivity threshold of 80%. The proposed system was proved to be non-inferior with the expert consensus on the MUSC dataset. Further, we developed an EEG classification system based on the output of the interictal pattern detector. Multiple features were derived from the CNN outputs, and the best features were selected based on p-values. The best features were then employed to train the EEG classifier. The proposed system achieved a cross-validation balanced accuracy of 81.61% on the MGH dataset. Later the system, trained on the MGH dataset was validated on 100 EEGs recorded at NUH. The epileptic EEG identification system achieved an accuracy of 88% for clinical validation on the NUH data.Doctor of Philosoph