Combination of complex-valued neural networks with silicon-loaded probes for millimeter-wave non-invasive blood glucose concentration estimation

Abstract

This paper proposes a millimeter-wave human glucose-concentration estimation system based on the combination of a complex-valued neural network (CVNN) and dielectric-loaded probes. The system observes the complex-valued scattering coefficients in the millimeter-wave transmission through a thin human tissue such as an earlobe and a finger web, and estimates the concentration value by utilizing the CVNN learning ability. In this paper, we demonstrate that the silicon loading at the probes enhances the CVNN ability, resulting in a better estimation in in vivo experiments. The results will lead to the realization of reliable and practical non-invasive human blood glucose monitoring systems in the near future.