| Summary: | This thesis is concerned with the analysis and development of methods for simultaneously distinguishing between non-ventricular rhythms, ventricular tachycardia and ventricular fibrillation in the electrocardiogram. A realistic experimental framework for assessing methods was developed that does not over-estimate ac- curacy of investigated methods, and descriptive statistics were used for reporting results of experimental simulations. The methods developed were tested against recent studies in the literature. The developed methods introduced high dimensional feature spaces for reducing the amount of information discarded, and the best method achieved 30% reduction in median error rates by combining multiple feature spaces, directly and in a hierarchical fashion, and through incorporation of rhythm context from past observations, as opposed to conducting analysis on the currently observed segment alone. The research conducted has not solved the problem of differentiating between non-ventricular rhythms, ventricular tachycardia and ventricular fibrillation entirely, and remains an open problem for research. Through the development of methods in this thesis and observations made, many more avenues are proposed for improving automated rhythm diagnosis in the electrocardiogram.
|
|---|
