Probabilistic pattern analysis to detect heart disease using electrocardiogram

Abstract

ECG describes the status of a heart as simple wave forms, and has important roles in clinical diagnosis. It is one of the frequently used methods in medical field. Computer-based ECG measuring and analyzing method for detecting heart disease has been used since the beginning of 1990s, and recently applying IT technology to these methods produces a faster and more exact method to identify heart diseases. This thesis aims to develop various methods to detect heart diseases using ECG. We propose a pattern analysis method of ECG features and an ontology creation method for heart disease classification. In order to demonstrate these methods, we implement an integrated ECG processing and heart disease detection system based on the pattern analysis method and the domain ontology method. For the pattern analysis, SVM-based pattern analysis is proposed to classify ECG beats and rhythms. For ontology creation method, we consider semantic relationships between classified beats and waveforms retrieved from ECG database, and propose the domain ontology creation method for heart disease detection. These methods are used in Physio-Grid project which provides a total medical environment for signal measurement, detection and diagnosis of heart diseases based on grid technologies.