Yesim Serinagaoglu

Application of Bayesian Methods to Electrocardiography

Date : December 12th, 2002

ABSTRACT:

The main motivation of inverse electrocardiography research is to recover the cardiac electrical activity in humans in as non--invasive a manner as possible. Due to attenuation and spatial smoothing that occurs in the thorax, inverse electrocardiography is an ill-posed inverse problem; small disturbances in the measurements lead to amplified errors in inverse solutions. Thus regularization methods or related techniques are required to stabilize the solution.

One basic limitation on performance of all inverse methods for ill--posed problems is the availability of good a priori information. Because of the great variability in how cardiac potentials behave, reconstruction of such potentials has thus far only had limited success. One possible way to achieve improved performance would be to acquire better information about an individual beat, and then try to incorporate this information into an inverse solution. A second limitation on successful use of inverse electrocardiographic solutions is the lack of a reliable evaluation tool that will enable researchers, and eventually clinicians, to obtain a performance analysis of the solutions obtained in a realistic setting.

In this thesis, we combine information from simulated body surface and venous catheter measurements, along with signals from previous epicardial recordings (training sets), and study whether this combination improves the inverse solution. We adopt a Bayesian estimation framework for the inverse problem to take advantage of Bayesian error metrics. We also propose that the confidence intervals obtained from Bayesian estimates are useful to interpret the results and to determine where on the heart we would expect more, or less, reliable estimation. The advantage of this error metric is that it does not require the knowledge of the true solution. We test our methods using torso and venous catheter measurements simulated from canine electrograms.

In addition to a Bayesian approach to the inverse problem, we propose a Bayesian multiresolution fiducial point estimation algorithm and presented preliminary results for this work using simulated and optical mapping data, provided by our collaborators.

Thesis Committee:
Prof. Dana H. Brooks (Superviser)
Prof. Hanoch Lev-Ari
Prof. Eric L. Miller
Prof. Robert S. MacLeod (University of Utah)
Dr. John Triedman (Boston Children's Hospital)