Enhanced physics modelling and multistatic data integration for breast tissue reconstruction in microwave radar imaging

Abstract

Microwave breast imaging offers a low-cost, portable, and non-ionizing alternative for breast cancer detection by exploiting the contrast in dielectric properties between malignant and healthy tissues. Its simplicity and safety make it particularly attractive for screening in under-served or remote communities where conventional imaging modalities may be unavailable. However, three core challenges have limited its diagnostic performance: (1) standard Delay-and-Sum (DAS) beamformers assume a homogeneous medium and ignore tissue heterogeneity; (2) they neglect the frequency dependence of dielectric properties; and (3) true propagation speeds in complex breast tissues are unknown and must be estimated. Moreover, existing image-quality assessments rely on single-pixel contrast and localization metrics, offering little guidance for algorithmic refinement. This thesis addresses these gaps through three methodologies. Firstly, an enhanced-physics beamforming algorithm was developed as a combination of an analytical binary-partitioning model with a full frequency-dependent propagation-speed formulation, yielding piecewise time delays. The enhanced physics modelling was observed to improve image quality. Secondly, a robust, sinogram-based boundary detection algorithm was proposed to extract realistic breast outlines from raw data from the Vector Network Analyzer (VNA), replacing idealized circular models and allowing for a boundary-aware beamforming and skin suppression for the differential imaging. Thirdly, a phase-based technique of experimental propagation speed extraction was developed using the multistatic microwave data.