Joint Inversion of Electromagnetic and Acoustic Data With Edge-Preserving Regularization for Breast Imaging

Abstract

Joint inversion of microwave and ultrasonic data for breast imaging is investigated with deterministic edge-preserving regularization by introducing auxiliary variables indicating whether a pixel is on an edge or not. These edge markers are shared by dielectric and acoustic parameters and are the link to fusion between modalities. They can be jointly optimized from the last parameter profiles of microwave and ultrasonic cases and guide the next optimization as coefficients of the regularization term. Alternate minimization is used to update acoustic contrast, edge markers and dielectric contrast. Comprehensive numerical experiments are carried out on breast phantoms, a simple synthetic one and three extracted from a database. The results show, with comparisons to more classical approaches involving total variation or cross-gradient regularization developed in parallel, that the joint inversion algorithm can gain from the high resolution of ultrasonic imaging and the high contrast of microwave imaging. The quality of microwave imaging is enhanced in clear fashion and small tumors detected.