Antenna Modeling Issues in Quantitative Image Reconstruction Using a Flexible Microwave Tomography System

Quantitative microwave imaging has been extensively studied in the past years as an alternative technique in biomedical imaging, with a strong potential in early stage breast can- cer detection [Keith D. Paulsen and Paul M. Meaney, Alternative Breast Imaging", The Springer International Series in Engineering and Computer Science, 778, 2005]. The image reconstruction involves a nonlinear inverse scattering problem, which consists to retrieve the dielectric prop- erties of the biological object from the measured scattered ¯eld, for an applied incident ¯eld. Consequently, the solution is highly sensitive to model errors in the incident ¯eld. This paper focus on the impact of this model error on the reconstructed quantitative image using a °exible robotic microwave imaging system, developed at MÄalardalen University, together with an itera- tive Newton-Kantorovich (NK) algorithm. This study is conducted during the development of the imaging system and the ¯rst quantitative images of a breast phantom are obtained. The robotic microwave imaging system is developed as a °exible experimental platform for biomedical imaging, where one of the applications is breast imaging. Using a robot controlled system the scattered ¯eld can be measured with a single transmitting/receiving antenna-pair, thus avoiding the mutual coupling that occur when an antenna-array is used. The scattered ¯eld is, herein, measured around a breast phantom, along a circular arc, in the horizontal plane with vertically polarized monopole antennas, considering a two dimensional transverse magnetic case (2D-TM). The radiated ¯eld from the transmitting antenna is modeled as a vertical polarized cylindrical wave in the numerical incident ¯eld model of the NK algorithm, where images with both the real- and imaginary permittivity pro¯le of the breast phantom are obtained. In this study, two di®erent monopole antenna designs are compared with the numerical incident ¯eld model. The di®erence between the antennas is the ground-plane design, where the ¯rst setup uses 4 wires forming a horizontal cross as a ground-plane. By varying the length of the wires and angel between the transmitting and receiving antenna the incident electromagnetic ¯eld is changed and could be modi¯ed to best ¯t the simulated ¯eld. The second setup uses a circular ground plane which will give a more rotational symmetric radiation pattern in the horizontal plane and a better match when comparing measured ¯elds with computed values. The comparison is done directly with the numerical incident ¯eld model, as well as the computed and measured scattered ¯eld, and ¯nally the impact on the reconstructed images by the NK algorithm are compared, using measured data from both antennas. The results show how the antenna selection impacts the error between the measured incident ¯eld and the numerical model, and how the quantitative image of an inhomogeneous object is a®ected by this model error.