Development of an intelligent Neurologist Support System (iNSS) for diagnosing epilepsy

Abstract

Epilepsy is a chronic neurological disease which is suffered by every one out of 100 people in the world. Neurologists are those types of clinicians which specializes in treatment if epilepsy. Even though this disease is very common but Neurologist in general is a rare commodity in the world. Hence in developing Countries like Pakistan their rarity is obvious. This work is an effort to supplement the clinicians during the diagnosis of Epilepsy. Electroencephalography is a very abstract way of analysing human’s brain activity. Voltage fluctuations are caused by the flow of ionic current in neurons. During electroencephalography these voltages are recorded using a voltmeter. Representation of these voltages is called as Electroencephalogram (EEG). Epileptic patient’s EEG exhibits some unique patterns which are called as Epileptic Patterns. Clinicians analyse these unique patterns to diagnose Epilepsy. The origin and location of these patterns also helps the clinician in diagnosing the type and intensity of the Epileptic disorder. In our work we have tried to supplement a clinician during the diagnostic analysis of an EEG of epilepsy suspected patient. Computer Assisted Analysis (CAA) tools are those software programs which helps the user in analysing, managing or organizing certain information in the form of a data. CAA has an immense amount of potential in assisting clinicians during EEG analysis. CAA tools highlights the epileptic patterns among whole of an EEG, which lessen up the fatigue for the clinician to analyse whole of the EEG hence saving his precious time. CAA is not built to replace the clinician rather they are there to assist them. Neurologists are the one who provide the ground truth to train these CAA tools so that they may classify the EEG as epileptic or benign. So, there is a need to develop a mechanism in the existing systems using which its incorrect markings can be mentioned and the system should improve its classification rate by learning from its mistakes. In this work we have developed a simple mechanism for clinicians to improve classification of the system while encountering any wrong markings by the system. The working of this system is based on application of discrete wavelet transform (DWT) on EEG epochs. The statistical features from the detailed coefficients representing the desired frequency ranges are then reduced using Principal Component Analysis which is then fed into a classifier. In this write up, first we have discussed our approach then we have shown the results. We have shown the performance of three types of classifiers i.e. Support Vector Machine (SVM), Quadratic Discriminant Analysis and Artificial Neural Network. We have found SVM to be the best working classification technique. Our work is an exhibition of importance and feasibility of a self-improving and user adapting computer assisted EEG analysis system for Epileptic diagnosis.