A Gauss-Markov mixture prior model for a variational Bayesian approach to microwave breast imaging

We deal with microwave imaging as a non-linear inverse scattering problem. The aim is to image an unknown object from measurements of the scattered field that results from its interaction with a known wave in the microwave frequency range. The modeling of the wave-object interaction is tackled through a domain integral representation of the electric field in a 2D-TM configuration. A variational Bayesian technique is used to solve the inverse problem, while prior information is introduced via a Gauss-Markov mixture model (MGM). The method is applied to microwave imaging for breast tumor detection. Inversion is performed from synthetic data obtained from a breast phantom built up from a MRI scan of a real breast and simulation results show the effectiveness of the method and emphasize the role of prior in the reconstruction improvement.