Microwave Imaging (MWI) is one of the most promising imaging modalities to emerge in recent years. It has been used in a range of applications, and one of the most notable application is medical imaging. Microwave medical imaging is based on the contrast between the dielectric properties of healthy and cancerous tissue. Compared to the existing imaging modalities including the widely used X-ray mammography, MWI is non-ionising, non-invasive and cost-effective. Among other medical imaging applications, MWI has been investigated to detect small tumours in the breast at a relatively early stage, and it has shown promising results in the detection of malignancies in a moderately dense human breast. The heterogeneity of healthy breast tissue, high computational cost of the MWI algorithms, and sensitivity of the microwaves to tissue boundaries are the major challenges in the development of a microwave based breast imaging system. This thesis focusses on the major challenges in MWI and proposes: (1) a novel pre- ltering technique to compensate for phase variations and frequency dependent loss that arise due to the dispersive nature of the biological tissue, (2) a massively parallel execution framework for microwave tomography to reduce the image reconstruction time, and (3) a multistage ultra wideband microwave tomography algorithm for quantitative breast imaging. Results indicate that the proposed pre- ltering improves the performance of conventional beamformers, enabling detection and accurate localization of multiple tumours in the dense human breast. The proposed multi-stage microwave tomography algorithm has been applied to quantitative imaging of the breast, and the performance of the proposed algorithm has been validated on anatomically and dielectrically realistic breast phantoms. In order to minimise the image reconstruction time, a single-instruction multiple-data based parallel execution framework has been developed, and a gain factor of more than 50 in the computational throughput has been achieved.
|Published - 2018