Ramdas Venkatachary

Automated Feature Extraction and Analysis of ST Segements from PTCA-BSPM Using the Wavelet Transform

August 26, 1998
2:00 PM
306 Egan

Abstract

The electrocardiogram (ECG) is a popular and useful diagnosis tool for many types of cardiac abnormalities. However, it contains only a very sparse spatial sampling of the distribution of electrical potentials of cardiac origin on the body surface. Therefore, significant information, such as spatially localized cardiac events, may not be detected by an ECG. In an effort to obtain a more thorough and inclusive understanding of the information contained in cardiac body surface potentials, researchers and some clinicians record the ECG with an extended number of leads (typically around 32---200). This procedure is known as Body Surface Potential Mapping (BSPM). The result is a large dataset consisting of a kind of `image sequence' of electrical potentials. Extracting physiologically relevant information from such a dataset is a non-trivial problem.

The Discrete Wavelet Transform (DWT) is a popular technique for multiresolution time-frequency analysis of a signal. With the choice of specific wavelets, the decomposition can be made time-invariant and linear phase across different stages of wavelet analysis. This is particularly useful for applications that require comparison of the wavelet coefficients across different stages of analysis.

The research work in this thesis focuses on the use of a specific variant of the DWT to identify and extract specific regions of interest from within BSPMs recorded during a medical procedure known as Percutaneous Transluminal Coronary Angioplasty (PTCA). In PTCA one of the arteries feeding the heart is occluded and the BSPM record consequent electrocardiographic changes. Here the goal is to determine the artery in which the PTCA occlusion was made based on the BSPM recordings. An algorithm for the automatic extraction of a specific region of interest in each lead will be presented. We will present results of applying our algorithm on patient data from two separate databases. In addition, we will also present results of using the algorithm to analyze a statistic of the information that is extracted in order to determine the occluded artery.

Thesis Committee:
Prof. D.H. Brooks (advisor)
Prof. V.K. Ingle
Prof. R. MacLeod, University of Utah