This thesis discusses the application of parallel computing in microwave tomography for detection and imaging of dielectric objects. The main focus is on microwave tomography with the use of a parallelized Finite Difference Time Domain (FDTD) forward solver in conjunction with non-linear stochastic optimization based inverse solvers. Because such solvers require very heavy computation, their investigation has been limited in favour of deterministic inverse solvers that make use of assumptions and approximations of the imaging target. Without the use of linearization assumptions, a non-linear stochastic microwave tomography system is able to resolve targets of arbitrary permittivity contrast profiles while avoiding convergence to local minima of the microwave tomography optimization space. This work is focused on ameliorating this computational load with the use of heavy parallelization. The presented microwave tomography system is capable of modelling complex, heterogeneous, and dispersive media using the Debye model. A detailed explanation of the dispersive FDTD is presented herein. The system uses scattered field data due to multiple excitation angles, frequencies, and observation angles in order to improve target resolution, reduce the ill-posedness of the microwave tomography inverse problem, and improve the accuracy of the complex permittivity profile of the imaging target.
The FDTD forward solver is parallelized with the use of the Common Unified Device Architecture (CUDA) programming model developed by NVIDIA corporation. In the forward solver, the time stepping of the fields are computed on a Graphics Processing Unit (GPU). In addition the inverse solver makes use of the Message Passing Interface (MPI) system to distribute computation across multiple work stations. The FDTD method was chosen due to its ease of parallelization using GPU computing, in addition to its ability to simulate wideband excitation signals during a single forward simulation.
We investigated the use of distributed Particle Swarm Optimization (PSO) and Differential Evolution (DE) methods in the inverse solver for this microwave tomography system. In these optimization algorithms, candidate solutions are farmed out to separate workstations to be evaluated. As fitness evaluations are returned asynchronously, the optimization algorithm updates the population of candidate solutions and gives new candidate solutions to be evaluated to open workstations. In this manner, we used a total of eight graphics processing units during optimization with minimal downtime.
Presented in this thesis is a microwave tomography algorithm that does not rely on linearization assumptions, capable of imaging a target in a reasonable amount of time for clinical applications. The proposed algorithm was tested using numerical phantoms that with material parameters similar to what one would find in normal or malignant human tissue.
","abstract_html":"<p>This thesis discusses the application of parallel computing in microwave tomography for detection and imaging of dielectric objects. The main focus is on microwave tomography with the use of a parallelized Finite Difference Time Domain (FDTD) forward solver in conjunction with non-linear stochastic optimization based inverse solvers. Because such solvers require very heavy computation, their investigation has been limited in favour of deterministic inverse solvers that make use of assumptions and approximations of the imaging target. Without the use of linearization assumptions, a non-linear stochastic microwave tomography system is able to resolve targets of arbitrary permittivity contrast profiles while avoiding convergence to local minima of the microwave tomography optimization space. This work is focused on ameliorating this computational load with the use of heavy parallelization. The presented microwave tomography system is capable of modelling complex, heterogeneous, and dispersive media using the Debye model. A detailed explanation of the dispersive FDTD is presented herein. The system uses scattered field data due to multiple excitation angles, frequencies, and observation angles in order to improve target resolution, reduce the ill-posedness of the microwave tomography inverse problem, and improve the accuracy of the complex permittivity profile of the imaging target. </p> <p>The FDTD forward solver is parallelized with the use of the Common Unified Device Architecture (CUDA) programming model developed by NVIDIA corporation. In the forward solver, the time stepping of the fields are computed on a Graphics Processing Unit (GPU). In addition the inverse solver makes use of the Message Passing Interface (MPI) system to distribute computation across multiple work stations. The FDTD method was chosen due to its ease of parallelization using GPU computing, in addition to its ability to simulate wideband excitation signals during a single forward simulation.</p> <p>We investigated the use of distributed Particle Swarm Optimization (PSO) and Differential Evolution (DE) methods in the inverse solver for this microwave tomography system. In these optimization algorithms, candidate solutions are farmed out to separate workstations to be evaluated. As fitness evaluations are returned asynchronously, the optimization algorithm updates the population of candidate solutions and gives new candidate solutions to be evaluated to open workstations. In this manner, we used a total of eight graphics processing units during optimization with minimal downtime.</p> <p>Presented in this thesis is a microwave tomography algorithm that does not rely on linearization assumptions, capable of imaging a target in a reasonable amount of time for clinical applications. The proposed algorithm was tested using numerical phantoms that with material parameters similar to what one would find in normal or malignant human tissue.</p>","abstract_has_math":false,"creators":["Holman, Michael William"],"institution":null,"degree_name":"Master of Science (MS)","degree_level":"Thesis","degree_discipline":"Electrical Engineering","degree_department":null,"school":null,"contributors":["Sima Noghanian"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2013,"date_issued":"2013-01-01T08:00:00Z","date_published":"2013-01-01T08:00:00Z","updated_at":"2026-07-24T03:26:24Z","subjects":["Breast Cancer, Evolutionary Algorithm, GPU Computing, High Performance Computing, Medical Imaging, Microwave Tomography"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://commons.und.edu/theses/1436","outbound_label":"Repository record","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Sima Noghanian"]},{"key":"dc:creator","label":"Author","values":["Holman, Michael William"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Thesis"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science (MS)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Breast Cancer, Evolutionary Algorithm, GPU Computing, High Performance Computing, Medical Imaging, Microwave Tomography"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://commons.und.edu/theses/1436"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["This thesis discusses the application of parallel computing in microwave tomography for detection and imaging of dielectric objects. The main focus is on microwave tomography with the use of a parallelized Finite Difference Time Domain (FDTD) forward solver in conjunction with non-linear stochastic optimization based inverse solvers. Because such solvers require very heavy computation, their investigation has been limited in favour of deterministic inverse solvers that make use of assumptions and approximations of the imaging target. Without the use of linearization assumptions, a non-linear stochastic microwave tomography system is able to resolve targets of arbitrary permittivity contrast profiles while avoiding convergence to local minima of the microwave tomography optimization space. This work is focused on ameliorating this computational load with the use of heavy parallelization. The presented microwave tomography system is capable of modelling complex, heterogeneous, and dispersive media using the Debye model. A detailed explanation of the dispersive FDTD is presented herein. The system uses scattered field data due to multiple excitation angles, frequencies, and observation angles in order to improve target resolution, reduce the ill-posedness of the microwave tomography inverse problem, and improve the accuracy of the complex permittivity profile of the imaging target.
The FDTD forward solver is parallelized with the use of the Common Unified Device Architecture (CUDA) programming model developed by NVIDIA corporation. In the forward solver, the time stepping of the fields are computed on a Graphics Processing Unit (GPU). In addition the inverse solver makes use of the Message Passing Interface (MPI) system to distribute computation across multiple work stations. The FDTD method was chosen due to its ease of parallelization using GPU computing, in addition to its ability to simulate wideband excitation signals during a single forward simulation.
We investigated the use of distributed Particle Swarm Optimization (PSO) and Differential Evolution (DE) methods in the inverse solver for this microwave tomography system. In these optimization algorithms, candidate solutions are farmed out to separate workstations to be evaluated. As fitness evaluations are returned asynchronously, the optimization algorithm updates the population of candidate solutions and gives new candidate solutions to be evaluated to open workstations. In this manner, we used a total of eight graphics processing units during optimization with minimal downtime.
Presented in this thesis is a microwave tomography algorithm that does not rely on linearization assumptions, capable of imaging a target in a reasonable amount of time for clinical applications. The proposed algorithm was tested using numerical phantoms that with material parameters similar to what one would find in normal or malignant human tissue.
"]},{"key":"dc:title","label":"Title","values":["Microwave Tomography Using Stochastic Optimization And High Performance Computing"]}]}],"canonical_facts":{"dc:contributor":["Sima Noghanian"],"dc:creator":["Holman, Michael William"],"dc:description.abstract":["This thesis discusses the application of parallel computing in microwave tomography for detection and imaging of dielectric objects. The main focus is on microwave tomography with the use of a parallelized Finite Difference Time Domain (FDTD) forward solver in conjunction with non-linear stochastic optimization based inverse solvers. Because such solvers require very heavy computation, their investigation has been limited in favour of deterministic inverse solvers that make use of assumptions and approximations of the imaging target. Without the use of linearization assumptions, a non-linear stochastic microwave tomography system is able to resolve targets of arbitrary permittivity contrast profiles while avoiding convergence to local minima of the microwave tomography optimization space. This work is focused on ameliorating this computational load with the use of heavy parallelization. The presented microwave tomography system is capable of modelling complex, heterogeneous, and dispersive media using the Debye model. A detailed explanation of the dispersive FDTD is presented herein. The system uses scattered field data due to multiple excitation angles, frequencies, and observation angles in order to improve target resolution, reduce the ill-posedness of the microwave tomography inverse problem, and improve the accuracy of the complex permittivity profile of the imaging target.
The FDTD forward solver is parallelized with the use of the Common Unified Device Architecture (CUDA) programming model developed by NVIDIA corporation. In the forward solver, the time stepping of the fields are computed on a Graphics Processing Unit (GPU). In addition the inverse solver makes use of the Message Passing Interface (MPI) system to distribute computation across multiple work stations. The FDTD method was chosen due to its ease of parallelization using GPU computing, in addition to its ability to simulate wideband excitation signals during a single forward simulation.
We investigated the use of distributed Particle Swarm Optimization (PSO) and Differential Evolution (DE) methods in the inverse solver for this microwave tomography system. In these optimization algorithms, candidate solutions are farmed out to separate workstations to be evaluated. As fitness evaluations are returned asynchronously, the optimization algorithm updates the population of candidate solutions and gives new candidate solutions to be evaluated to open workstations. In this manner, we used a total of eight graphics processing units during optimization with minimal downtime.
Presented in this thesis is a microwave tomography algorithm that does not rely on linearization assumptions, capable of imaging a target in a reasonable amount of time for clinical applications. The proposed algorithm was tested using numerical phantoms that with material parameters similar to what one would find in normal or malignant human tissue.
"],"dc:identifier":["https://commons.und.edu/theses/1436"],"dc:subject":["Breast Cancer, Evolutionary Algorithm, GPU Computing, High Performance Computing, Medical Imaging, Microwave Tomography"],"dc:title":["Microwave Tomography Using Stochastic Optimization And High Performance Computing"],"thesis:degree_discipline":["Electrical Engineering"],"thesis:degree_level":["Thesis"],"thesis:degree_name":["Master of Science (MS)"]},"updated_at":"2026-07-24T03:26:24Z"}