Strategies for improving the image reconstruction in microwave tomography

Abstract

Microwave tomography is a technique for imaging dielectric properties. It has gained increasing attention during last decades, in particular by its potential application to detect breast cancer. The purpose of this thesis is to investigate different optimization methods in the image reconstruction procedure. The conjugate gradient (CJG), Gauss-Newton (GN), Quasi-Newton (QN) and One step secant (OSS) method have been tested and compared. The result is that the conjugate gradient method (CJG) is the best choice among the investigated methods. Furthermore a method is proposed for making use of a priori data based on knowledge about the dielectric properties of the investigated object. In numerical tests this method has proven to enhance the convergence rate. A method for choosing the step length in each iteration is also proposed that requires approximately 55% less calculation time for a typical computation scenario compared to a successive parabolic interpolation. Moreover an interrogating pulse with Gaussian frequency distribution is compared to a pulse with more uniformly distributed energy. Using the latter in a numerical simulation gives a better image reconstruction in the first few iterations. An analysis of experimental data is also made as an attempt of detecting a real tumor. It turns out to be impossible and possible sources of error are identified.