Thesis title: Towards Deep Learning Enhanced Microwave Medical Imaging
This PhD thesis is devoted to the thriving field of deep learning enhanced microwave imaging. Microwave imaging is a technology with applications in different fields involving the inspection of buried or nested objects. However, microwave imaging faces several challenges that are slowing down its adoption in some areas. In particular, the mathematical problem at the core of microwave imaging belongs to the class of inverse scattering problems, which are nonlinear and ill-posed. Recently, employing deep learning techniques to address the difficulties of inverse scattering problems has received a lot of attention. In fact, the capability of deep learning techniques to solve complex problems makes them a potential tool to reliably solve the inverse scattering problem. In this regard, deep learning approaches to microwave imaging are grouped in 3 families, each tackling the inverse scattering problem in a different way. Along the thesis, one approach belonging to each family is discussed. The content of the thesis is organized in two separate research activities. Microwave imaging is of relevance in biomedical applications, where it is considered an emerging non-ionizing, low-cost and portable imaging modality. For this reason, the first research activity presents two examples of deep learning microwave medical imaging: monitoring of hyperthermia treatment and brain imaging. In both examples, the a posteriori performance assessment shows compelling capabilities. In the second research activity, a physics-assisted framework is proposed. The key aspect of this framework is tackling the inverse scattering problem from a more general perspective. To this end, three framework implementations are discussed, each solving a different imaging objective. Performance assessments as well as validation on experimental data are provided to confirm the applicability of the framework. To elaborate further on the applicability, the research activity concludes with a successful implementation of the framework in a biomedical application: brain stroke imaging.