Identification of ECG anomalies through deep deterministic learning / Uzair Iqbal

• Main Author: Uzair , Iqbal
• Format: Thesis
• Published: 2020
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://studentsrepo.um.edu.my/14407/2/Uzair.pdf; http://studentsrepo.um.edu.my/14407/1/Uzair_Iqbal.pdf; http://studentsrepo.um.edu.my/14407/

Electrocardiography (ECG) is a primary diagnostic tool for measuring the malfunctioning of the heart muscles in the context of morbidity of different cardiac diseases and arrhythmia. Different existing techniques and methods delivered accurate cardiac diseases myocardial infarction (heart stroke) and atrial fibrillation recognition. However, there are still some flaws in existing methods like recognition of special myocardial infarction situation flattened T wave in “Non-Specific ST-T Changes (nsst-t)” and reduction of computational cost in cardiac diseases recognition. Accurate recognition of cardiac diseases along with least computational complexity and feature analysis of flattened T wave in myocardial infarction remains an open job. In this research, three different datasets were used for experimental activities. Two datasets are publicly available namely; Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) arrhythmia database and Physikalisch-Technische Bundesanstalt (PTB), and the third dataset are exclusively obtained from the University of Malaya Medical Center (UMMC), Kuala Lumpur, Malaysia. This thesis presents the major contributions in perspective of, accurate as well as least computational complex in recognition of atrial fibrillation and flattened T wave situation in myocardial infarction detection and prediction. Two new deterministic methods are proposed namely; deep deterministic learning (DDL) and model driven deep deterministic learning (MDDDL) which delivered impressive results in recognition and predictive classification of atrial fibrillation and flattened T wave situation in myocardial infarction (i.e., ≤99.97%). Finally, both the proposed models DDL and MDDDL are further useful for recognition and predictive classification of the other malfunctions of the heart.