FUSION OF ECG/EEG FOR IMPROVED AUTOMATIC SEIZURE DETECTION USING DEMPSTER-SHAFER THEORY OF EVIDENCE

FUSION OF ECG/EEG FOR IMPROVED AUTOMATIC SEIZURE DETECTION USING DEMPSTER-SHAFER THEORY OF EVIDENCE. Masters thesis, King Fahd University of Petroleum and Minerals.

[img]
Preview
PDF
finaldraft.pdf - Submitted Version

Download (3MB) | Preview

Arabic Abstract

تهدف الدراسة إلى تطوير طريقة جديدة للتحليل المشترك للتخطيط الكهربائي للقلب (ECG) والتخطيطات الكهربائية للدماغ (EEG). وتعتمد هذه الطريقة على مبادئ الأدلة النظرية لدمستر وشافر(DS).ويعتمد تحليل التخطيطات الكهربائية للدماغ(EEG) على طريقة الزمن و التردد (time-frequency) للتعرف عن النوبات القصيرة وذلك باستخراج سمات مميزة من هذا التحليل. أما فيما يخص تحليل تخيط القلب(ECG) فنقترح استعمال طريقة المويجات (wavelets). وهذا التحليل يؤدي إلى استخراج عدة سمات نذكر منها فاصل RR، PR،QR إلى غيرذلك. ونذكر أن في كل من الحالتين نستعمل طريقة التحليل التميزي الخطي(LDA) لتصنيف الإشارات إلى إشارات عادية أو إشارات نوبات مرضية. وللتحسين من أداء النظام المقترح، قدمنا طريقة مزج EEG و ECGباستعمال نظرية DS والتي أدت إلى تحسين أداء النظام في تحديد الزمن والتعرف على النوبات الدماغية بنسبة تفوق 97%.

English Abstract

Objective: A Dempster Shafer based combination method is presented for the seizure detection algorithm using Electroencephalogram (EEG) and Electrocardiogram (ECG). The individual results from the EEG and ECG are improved using this combination method. EEG algorithm: A time frequency (TF) based seizure detection algorithm is presented. The proposed technique uses features extracted from the Singular Value Decomposition (SVD) of the TF representation of EEG. These features are used with a simple Linear Discrimination Analysis (LDA) for classification of EEG traces into seizure and non seizure activity. A seizure classification accuracy was achieved outperforming most existing algorithms. ECG algorithm: A seizure detection technique which fully utilizes the ECG wave by extracting all the features which are found to be effected during a seizures is presented. In the previous approaches focus was only placed on the RR duration but none of the researches focused on the other features of an ECG wave which are affected during a seizure. In our research we included RR mean, RR variance, QT duration, PR duration, P wave height and variance as the features to train Linear Discriminant Analysis (LDA). These features are found to be different for a healthy and a seizure affected individual in the literature. The results showed a classification accuracy which outperform the previous seizure detection techniques. Combination: Dempster Shafer rule is used for combination of the above two algorithm. The combined classification accuracy obtained outperforms any existing seizure detection algorithms.

Item Type: Thesis (Masters)
Subjects: Engineering
Research
Electrical
Department: College of Engineering and Physics > Electrical Engineering
Committee Advisor: A. Deriche, Mohamed
Committee Co-Advisor: Mohandes, Mohamed
Committee Members: Zidouri, Abdelmalek and Maalej, Nabil and Arafat, Samer M.
Depositing User: Mohammed Siddiqui (g200804960)
Date Deposited: 10 Oct 2011 10:35
Last Modified: 01 Nov 2019 15:30
URI: http://eprints.kfupm.edu.sa/id/eprint/138533