Simulation Study on the Determination of Cardiac Ejection Fraction by Electrical Impedance Tomography Using a Hybrid Heuristic Approach

Cardiac Ejection Fraction (EF) is an important parameter to analyze the efficiency of the heart as a pump, and it is clinically correlated to the functional status of the organ. Electrical impedance tomography (EIT) can be applied to measure EF. Its main advantage is the possibility of frequently monitoring Ejection Fraction, in contrast with other non invasive methods like Computer Tomography, Magnetic Resonance, among others. In addition, EIT present low cost and high portability, other features that justify the research for solutions involving this technique to monitor EF. The goal of EIT is to reconstruct images of the conductivity distribution of the interior of a conductor domain. It can be accomplished by applying electric currents and measuring the electrical potential obtained on the boundary of the body. Such problem can be mathematically classified as a non-linear inverse problem and can be addressed as an optimization problem. This work addresses the problem from both an optimization and a simulation point of view, using a two-dimensional model. It is proposed a sequential and a parallel version of a method for estimation of EF based on a hybrid heuristic composed by a Levenberg-Marquardt Method (LMM) as local search inside an Iterated Local Search (ILS) scheme. We also propose a parallel version of such technique. The final objective of the proposed technique is not to recover images themselves from the inside of the domain for diagnostic purposes, but to investigate aspects of a portable and relatively cheap method to continuously monitor EF. Thus, to determine an initial approximation for the cavities of a patient, our methodology assumes the existence of recent images of the heart with enough spatial resolution. The boundaries of the ventricular cavities are then extracted from these images and parameterized. The next step is the solution of an EIT inverse problem. This parameterization of the heart geometry allow the reduction of the search space, reducing, therefore, the computational cost of the proposed method. Experiments are performed on synthetically generated data for electric potentials. Preliminary results comparing the LMM and the new hybrid proposed approach, as well as its parallel version, are presented and discussed.