A Hybrid Simulation-Measurement Calibration Scheme for Microwave Brain Imaging

This thesis assesses an innovative hybrid technique to calibrate a microwave imaging (MWI) system combining synthetically emulated measured data with pre-simulated scenarios while retrieving a brain stroke condition. Stroke is a medical emergency that occurs when a vessel within the brain bursts or clots, causing the death of millions of cells per minute and requiring prompt diagnosis and treatment. Currently, clinicians support their appraisal on well-established image-based technologies such as magnetic resonance imaging (MRI) and computed tomography (CT). However, novel complementary technologies like MWI, a harmlessness, cost-effective, and non-bulky alternative, emerged in the last years. The calibration proposed here, attempts to ponder variations between the realized MWI antenna array elements caused by unavoidable manufacturing tolerances and placement compared to the nominal modeled electromagnetic (EM) scenario. The proposed method aims to improve the inversion mechanism's reliability applied by a general MWI algorithm, which deals with non-linearity and ill-posedness and yields a recovered dielectric contrast map given the scattered field sampled through the antenna array. Hence, the approach requires an initial parametric assessment of the device under test by running a large set of numerical simulations accounting for possible physical and electrical mismatches compared to the nominal case. Then, the collected data has been processed via the singular values decomposition (SVD), obtaining the minimal set of independent vectors (basis) that maximize the amount of information. Thus, each antenna response (here synthetically emulated) is projected on the generated basis functions, according its behaviour, in terms of near field and scattering parameters, to the real system under calibration. The EM model considers accurate numerical a priori information on the system (geometry, materials, working conditions) and provided reliable EM behaviour, which is essential to achieve adequate imaging performance. To this end, the modeling considers a geometrical component followed by an EM one. The first part approaches a realistic characterization of the MWI device via a high-fidelity CAD model, including in the considered MWI system twenty-four monopole antennas enclosed each in a semi-solid coupling medium and placed conformally to an anthropomorphic head phantom. In the second part, an in-house EM finite-element-method-based solver computes the system's scattering parameters and EM fields. The calibration scheme has been validated on synthetic cases, which simulated the hemorrhagic stroke events at different stages by varying the stroke dimensions. Highly perturbed scenarios are assessed, comparing the reconstructed images before and after the calibration. The procedure has proven to enhance stroke detection capability, providing a promising solution for MWI systems, generalizable to different imaging algorithms. It is planned to extend the evaluation by applying the scheme to actual measured data in future work.