91ɬÂþ

Abstract

Breast cancer remains one of the leading causes of mortality among women, and early detection is essential for improving survival rates. X-ray mammography, the current gold standard, exposes patients to ionizing radiation that limits exam frequency, while magnetic resonance imaging is costly and reserved for high-risk cases. Microwave imaging uses non-ionizing radiation and exploits dielectric contrast between healthy and malignant tissues, making it an attractive low-cost complement to established diagnostics. Over the last decade, the RF Breast Cancer Detection Group at 91ɬÂþ has been developing flexible near-field biosensor arrays mounted on bra-like supports with the long-term vision of a comfortable device suitable for frequent screening.

Building on this platform, my work first redesigned the sensing array to improve near-field performance. I developed several generations of 3–5 GHz PDMS-based near-field antennas, assembled into arrays of up to 16 elements with hexagonal enclosures to concentrate the electromagnetic fields inside the tissue. Full-wave simulations and measurements showed that the new array suppressed surface-wave crosstalk between elements and reduced back radiation compared ith the legacy polyimide stepped-monopole sensor. Integrated into a hemispherical structure, the antennas were validated over carbon-based breast phantoms replicating the dielectric properties of fat, glandular tissue, skin, and tumour inclusions.

To eliminate reliance on expensive and bulky laboratory setups, such as a vector network analyzer (VNA) or custom ultrawideband (UWB) electronics normally used with these arrays, I developed a low-cost software-defined radio (SDR) RF transceiver system based on the open-source USRP platform. The complete system was characterized in terms of RF performance and showed results comparable to the VNA in the frequency range and power levels of interest. A denoising autoencoder neural network was then trained using VNA data as a reference to reduce the remaining performance and stability gap by suppressing unwanted spectral artefacts and trends in the SDR signals.

On the image reconstruction side, after revisiting conventional time-domain and frequency-domain beamformers facilitated by the Chirp-Z transform (CZT), I introduced a new full-wave confocal beamformer that embeds simulated near-field Green’s functions of the PDMS sensor array. Using full-wave HFSS models and measured data, I benchmarked these beamformers with standard image quality metrics. This framework allowed me to assess the performance of the newly designed system against the established 91ɬÂþ UWB prototype, showing that it preserved imaging performance at a fraction of the hardware cost and size.

Lastly, this thesis extends the microwave scattering principles used for breast screening to the problem of skin cancer screening, with the goal of establishing a numerical method for skin lesion characterization. The proposed approach uses a pair of Vivaldi antennas specifically designed to propagate surface waves along the tissue–air boundary, from which the effective permittivity of skin lesions is estimated. The underlying theory was validated through full-wave simulations, and three different Vivaldi prototypes were designed and fabricated to explore practical realizations.