Fast synthesis of breast monostatic microwave response for massive dataset generation

Abstract

Microwave (MW) technology has been researched for the past decades as a promising technique for breast cancer detection. However, its value for the medical community compared to existing techniques is still undetermined. One of the reasons is the lack of large enough representative datasets of MW signals for evaluation and improvement of the developed signal processing algorithms. To overcome this, we present a fast hybrid method combining geometrical optics, physical optics and full-wave analysis that can model the direct problem, allowing us to create digital twins of experimental microwave monostatic setups, for fast generation of large representative datasets. For validation purposes, a digital twin of our in-house experimental dry microwave setup was implemented – MWSensing – and made openly available to the interested community. We have tested the model against experimental results in the 2-5 GHz range for cylindrical, elliptical and anthropomorphic, homogeneous and heterogeneous breasts phantoms, with a tumor placed inside, in different positions. We show that the MW signals obtained from MWSensing and from the experimental setup are of the same nature and provide similar imaging results. With this framework, it becomes possible to build a large dataset of MW signals, quickly and efficiently, enabling users to evaluate firsthand their algorithms before moving on to more advanced test stages.