Microwave medical imaging of the human neck using a neural-networks-based inversion procedure: A phantom study

A preliminary experimental validation of a machine-learning technique for microwave imaging of the neck is reported in this paper. Specifically, a fully-connected neural network is used to retrieve the dielectric properties of a simplified neck phantom. The architecture of the network, e.g., number of layers and neurons in each layer, is optimized through a numerical analysis on simulated data. The initial results confirm that it is possible to train the network with simulated data only and to test it with real data, obtaining good reconstruction results.