Kadagattur Srinidhi

Convex Optimization Algorithms for Inverse Electrocardiography with Physiologically Relevant Constraints

Monday, August 23, 1999
11:00 AM
406 Egan

Abstract

The goal of solutions to the inverse problem of electrocardiography is to characterize normal and abnormal activity of the heart based on measurements of electrical potential on the surface of the body. The inverse problem is non-trivial because the smoothing and the attenuation inside the volume conductor make the problem ill-posed. Thus regularizing constraints are required to stabilize the inverse solution. In this thesis, we incorporate two new classes of constraints, one spatial and one temporal, into a formulation of the inverse problem based on an admissible solution approach and using the ellipsoid algorithm. The spatial constraint is designed to more accurately, in comparison to previously reported techniques, reconstruct the shape and amplitude of the steep activation wavefronts which are the macroscopic driving sources of cardiac electrical activity. The approach we take is modeled on edge-preserving methods for image restoration but uses a different formulation of the both the problem and the constraint based on an admissible solution method using the ellipsoid algorithm previously developed for inverse electrocardiography in our lab. The temporal constraints are designed to take advantage of prior knowledge that the sources are temporally correlated although the physics of the inverse problem are quasi-static. The use of temporal constraints significantly increases the problem size and, hence, the computational complexity of inverse solutions. In this thesis, we also describe a distributed computing testbed we have implemented with efficient load sharing to handle this increased computations.

Thesis Committee:
Prof. Dana H. Brooks (advisor)
Prof. Eric L. Miller
Prof. Davod R. Kaeli