Development of Novel Algorithms for Microwave Medical Imaging Applications

Student thesis: Doctoral ThesisDoctor of Philosophy


Non-ionising microwave (MW) is used in microwave imaging (MWI) to recover the information of the investigated tissues, such as the breast and head, for medical applications. Unlike radar-based MWI, microwave tomography (MWT) aims to reconstruct the dielectric properties by solving an electromagnetic (EM) inverse scattering problem (ISP) with measurement data obtained by an antenna array that transmits and receives MW signals. This requires a robust and efficient algorithm, as the inverse problem is ill-posed and highly nonlinear.

In this thesis, a novel distorted Born iterative method (DBIM) algorithm is proposed and applied to different scenarios, mainly for head imaging cases with the finite element method (FEM) and finite difference time domain (FDTD) method used as forward solvers to simulate wave propagation. The fast iterative shrinkage/thresholding algorithm (FISTA) is used as an inverse solver to solve the resulting linear systems. A novel two-dimensional (2-D) FEM-based forward solver that provides great efficiency is developed for DBIM, which utilises the FEM matrix to build the DBIM matrix without the need for interpolation. The 2-D FDTD-based DBIM approach has been extended to three-dimensional (3-D) versions, and an in-house 3-D FDTD forward solver is implemented with graphics processing unit (GPU) acceleration for the DBIM. With the help of the compute unified device architecture (CUDA) toolkit and MEX functions in Matlab, the 3-D implementation can use the high performance of GPU and Matlab’s capacity for matrix computation directly without any interface problem.

The proposed FEM-based and FDTD-based DBIM approaches are combined with FISTA to obtain better reconstruction performance. The advanced inverse solver FISTA uses a shrinkage operator and accelerated Nesterov momentum to solve ill-posed linear systems with improved robustness and efficiency, which has a better performance compared to traditional inverse solvers such as gradient descent (GD) type methods.

Two efficient tools are proposed to improve the DBIM reconstruction results. The first tool is a preprocessing technique that employs the time gating technique to denoise the experimental signals, which can improve the signal quality and thus improve reconstruction results. The second tool is a postprocessing technique using k-means to cluster and classify the obtained reconstruction values, helping to better distinguish targets from the background.

These proposed algorithms are validated with numerical data and experimental data of different scenarios, including basic cylindrical models and complex models such as head models. A preliminary study of the reconstruction possibility of head imaging by the proposed FDTD-based DBIM-FISTA approach is presented, including an anatomically complex head phantom, the Zubal head phantom.

The Zubal head phantom is a 3-D magnetic resonance imaging (MRI)-derived voxel-based anthropomorphic phantom of human males from Yale University, which is suitable for testing and evaluating computer-based modelling and simulation calculations. In this thesis, it is converted to a simplified multi-layer phantom and used in different imaging scenarios to investigate MWT head/brain imaging with limited information and offset antenna rings.
Date of Award1 Feb 2023
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorPanos Kosmas (Supervisor) & Zoran Cvetkovic (Supervisor)

Cite this