The investigation of the effect of phase reversal and slow wave detection of ecg artifact in EEG signals

Electroencephalography (EEG) is a technique that allows non-invasive acquisition and visualization of electrical activities produced by the brain. With this technique, it is possible to evaluate brain signals, detect neurological disorders and certain conditions in the nervous system. On the other hand, EEG signals (EEGs) may also contain unwanted signals (artifacts) originating from outside the brain. It takes a long time to analyze and interpret the large amount of information in long-term EEGs. However, the duration of examination and interpretation of artifact-contaminated EEGs increases further, causing possible epileptiform activities to be overlooked. For this reason, automatic systems are needed to remove artifacts in EEGs. One of the artifacts observed in EEG signals is the Electrocardiogram (ECG) signal. ECG signals have a regular pattern and can be confused with spikes and sharp waves in terms of their morphology. Spikes and sharp waves that can be observed at regular intervals in EEGs cannot always be considered as epileptiform activity. Therefore, in this thesis, it is aimed to eliminate ECG artifacts in EEGs with high accuracy. For this, Zero Phase Filtering (ZPF) and Normalized Least Mean Square Filtering (NLMS) methods were used for ECG artifacts removal. The effects of ECG artifacts on the EEGs were investigated using some epileptiform activities. The first of the epileptiform activities examined is Phase Reversal (PR). PR, which can be observed in bipolar montages, is an indicator of the location of the maximum voltage and is very important especially in the determination of the localization of focal epilepsies. The other epileptiform activity examined in the study is Slow Wave (SW). Since EEGs in the ictal period are usually observed rhythmically, SW becomes an important sign in such cases. Peak detection (iPeak) and MinPeakProminence (MPP) methods were used for the detection of PR and SW, which are discussed in this thesis, respectively. In the performance evaluation of all methods in the study, classifier metrics (such as confusion matrix, accuracy) were used. The thesis basically consists of two parts. The first part consists of detecting the PR and SW in the raw EEGs without artifact removal, and the second part consists of detecting the PR and SW in the EEGs obtained after removing the ECG artifact with filtering. In the evaluation of the performances of the applied methods, the EEG dataset obtained retrospectively from the Neurology Department of the Selcuk University Medical Faculty Hospital (SUH), the dataset from the Boston Children's Hospital (CHB-MIT) and the dataset from the Bonn University Epileptology Department (BONN), which is open to the public, were used. In this thesis study, in the experiments conducted to examine the effects of ECG artifacts on the PR and SW in the EEGs, the accuracy rates obtained in the detection of the PR and SW in the raw EEGs in the SÜH data set were determined as 83.1% and 95.5%, respectively. The accuracy rates obtained after the removal of the ECG artifacts were 88.97% for the detection of PR and 95.3% for the detection of SW as a result of ZPF; As a result of NLMS filtering, it was found to be 86.29% for the detection of PR and 95.2% for the detection of SW. When the obtained accuracy rates were examined, it was observed that ECG artifact removal increased the success in the detection of PR, but did not have a great effect on the success in the detection of SW. As it is known, ECG signals are signals with high amplitude and therefore they can hide signals that may be PR. The reason for the increase in the success in detecting PR in the study is the emergence of PR with the elimination of high amplitude ECG artifacts. On the other hand, SW is waves with low amplitude and frequency. Therefore, after removing the high amplitude ECG artifact, no significant difference was observed in the success of SW detection. As a result, this thesis study contributed to the detection of PR in the EEGs with ECG artifact removal. In a future study, the effects of ECG artifacts on the detection of spikes and sharp waves can be investigated.