Electrical Parameters of Biological Cells from Dielectric Spectra - An Inverse Mapping Solved by Machine Learning

Machine learning techniques are currently used extensively in many aspects of science and technology. We apply these techniques to directly estimate electrical parameters of a suspension of biological cells from their dielectric response in radiofrequency. Due to its advantageous properties, a “broad learning” neural network architecture is then employed to realize the inverse mapping. The inversion results show that only the electrical conductivity of the cell membrane and the dielectric constant of the outer medium have deficient reconstruction accuracy, due to their singularities against the direct data; a possible further informed regularization is proposed to circumvent this problem.