Recovery of Prior Information for Breast Microwave Imaging Using Neural Networks

A recently developed neural network architecture for recovering the radius, height, and bulk complex-valued permittivity of the fibroglandular region of a human breast model from microwave measurements is extended to multiple frequencies. Results are presented for synthetic models with different sized fibroglandular regions both with and without a tumor present. The performance of this neural network architecture for single- and multi-frequency data in the 1.1 - 1.5 GHz range is demonstrated. Both neural networks are able to recover the desired bulk parameters of the fibroglandular region, with multi-frequency data leading to improved fibroglandular property estimates.