A 2-stage Gauss-Newton reconstruction technique for improved object detection in microwave imaging

We have developed a 2-stage Gauss-Newton iterative reconstruction technique to improve the general image quality with our microwave tomographic imaging system. It has been applied to simulation, phantom and in vivo breast imaging experiments with quantifiable improvement in the recovered property distributions over those achieved with our original, Levenberg-Marquardt approach. The latter method incorporates spatial filtering at the end of each iteration as a means of stabilizing the reconstruction process. However, the stabilization resulted in blurring of fine structures within the imaging region, particularly in the case of small high contrast objects. The new technique utilizes the original approach as the first stage in a 2-step strategy, allowing the algorithm to identify a distribution in the neighborhood of the ideal solution without converging to an unwanted local minimum. This intermediate result is subsequently employed as an initial estimate for the second stage process-a Tikhonov-based reconstruction with a weighted Euclidean distance penalty term to constrain the final result to be within the vicinity of the intermediate solution while allowing full impact of the electric field minimization without the detrimental effects of the spatial smoothing of the first step. While this approach has enhanced the recovery of features internal to the breast during microwave examinations, it has also improved our capability to recover small, high-contrast objects. This paper illustrates quantifiable improvements in the recovery of small objects in a laboratory setting.