Performance assessment of the DBIM-TwIST algorithm in three-dimensional microwave imaging problems

Student thesis: Doctoral ThesisDoctor of Philosophy


Microwave tomography (MT) aims to recover the dielectric profile of a domain by solving an inverse problem. Sophisticated algorithms and hardware are necessary for solving this problem. This thesis employs a novel distorted Born iterative method (DBIM) and the two-step iterative shrinkage thresholding (TwIST) implementation for reconstructing the complex permittivity of both two-dimensional (2-D) and three-dimensional (3-D) MT problems. Optimization approaches in the algorithm, as well as improvement in the experimental measurement process, are applied to improve the performance of reconstructions of the algorithm.
Moreover, this thesis presents preliminary results from a three-dimensional (3-D) DBIM-TwIST algorithm to deal with the ill-posed 3-D problem effectively. Satisfactory reconstructed images are obtained after applying to the 3-D DBIM-TwIST algorithm techniques such as frequency-hopping, 푳ퟏ norm regularization, and optimization of the initial guess. The results demonstrate that these techniques are necessary in order to reconstruct the breast interior in the 3-D microwave imaging problem. This impact is quantified by calculating and plotting relative reconstruction errors for various scenarios.
Furthermore, the thesis presents the resolution capabilities of adaptive thresholding methods in the context of iterative microwave imaging (MWI) algorithms. The simulation cases involve two closely located cylindrical targets of high dielectric contrast with respect to the background in an MT setup simulated in CST. To this end, we apply the respect to the background in an MT setup simulated in CST. To this end, we apply theDBIM algorithm and compare the TwIST implementation with a conventional conjugate gradient least squares (CGLS) method as linear solvers at each DBIM iteration. Our results demonstrate that applying the TwIST approach can resolve the two closely-locatedtargets much more accurately than other linear solvers such as the CGLS, under identical settings in the DBIM algorithm.At last, the thesis presents 3-D microwave imaging (MWI) reconstructions from a wideband microwave tomography system with potential application to medical imaging.The system is fully immersed in coupling liquids such as glycerine/water solutions. Weapply the DBIM-TwIST algorithm to solve both 2-D and 3-D microwave imaging problems of a cylindrical target in a homogeneous background, based on the data measured from an experimental system as well as simulated from its CAD model via CST Microwave Studio. We also present 3-D reconstructions for experiments with two cylinders filled with different materials inside the tank where the antennas are immersed.Results in all cases show reasonably good performance for the 3-D algorithm, which manages to detect and locate the target in all three dimensions, despite the increased computational complexity and increased ill-posedness of the 3-D MWI problem compared to its 2-D counterpart.
Date of Award1 Mar 2021
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorPanos Kosmas (Supervisor)

Cite this