JultikaUniversity of Oulu repository

Abstract

Epilepsy is a prevalent neurological disorder that affects millions of people globally, and its diagnosis typically involves laborious manual inspection of electroencephalography (EEG) data. Automated detection of epileptic seizures in EEG signals could potentially improve diagnostic accuracy and reduce diagnosis time, but there should be special attention to the number of false alarms to reduce unnecessary treatments and costs. This research presents a study on the use of machine learning techniques for EEG seizure detection with the aim of investigating the effectiveness of different algorithms in terms of high sensitivity and low false alarm rates for feature extraction, selection, pre-processing, classification, and post-processing in designing a medical device for detecting seizure activity in EEG data. The current state-of-the-art methods which are validated clinically using large amounts of data are introduced.

The study focuses on finding potential machine learning methods, considering KNN, SVM, decision trees and, Random forests, and compares their performance on the task of seizure detection using features introduced in the literature. Also using ensemble methods namely, bootstrapping and majority voting techniques we achieved a sensitivity of 0.80 and FAR/h of 2.10, accuracy of 97.1% and specificity of 98.2%. Overall, the findings of this study can be useful for developing more accurate and efficient algorithms for EEG seizure detection medical device, which can contribute to the early diagnosis and treatment of epilepsy in the intensive care unit for critically ill patients.