Machine Learning based Reconstruction of Point-Like Scatterers in a Portable Microwave Detection Device

Access to breast cancer screening is limited in low-income and remote areas, resulting in late-stage diagnosis and increased mortality rates. A portable microwave system was created by minimizing the device cost, size, and complexity. The small, cylindrical device features twenty-six patch antennas and inexpensive vector network analyzers operating from 0.7 - 3 GHz. A microwave radar model was modified to simulate S-11 measurements of the physical device. Radar simulations were performed on numerical phantoms consisting of two rod-like point scatterers with varying reflectivities of 10%, 30%, 50%, 70%, 90%, or 100%. A convolutional neural network (CNN) was trained to directly reconstruct the rod phantoms from their simulated S-11 sinograms. Despite the narrow bandwidth, the CNN could detect point scatterers with an accuracy up to 85%, improving on conventional resolving capabilities. Artificial intelligence microwave sensing methods offer promising possibilities for automated, low-cost breast cancer screening.