An optimized GPU-accelerated FDTD method for microwave imaging using a fast nonlinear inverse scattering algorithm

Summary form only given: Microwave imaging has received considerable attention as a low cost, non-invasive, non-ionizing method for breast cancer detection. In previous work, we have presented a time-domain nonlinear inverse scattering algorithm with multiparameter optimization for microwave imaging. In order to apply this algorithm to an experimental system that we have developed, it is crucial to have an accurate forward model of the imaging cavity. In this presentation, the modeling of the cavity using an in-house GPU accelerated finite-difference-time-domain (FDTD) method will be introduced, demonstrating several optimizations for increased computational efficiency and accuracy. S-parameter simulations of the cavity antennas comparing the results with the commercial software packages Ansys HFSS and CST MWS will be shown. Finally, results from microwave imaging tests of our GPU accelerated inversion algorithm using this fast forward model for both breast cancer detection and for real-time thermal monitoring of focused hyperthermia will be presented.The imaging cavity is a dodecagon consisting of 12 panels. Each panel has three dual-band bow-tie patch antennas operating at 915MHZ and 2.1GHz. In order to accurately capture the fine geometry of the cavity, we have utilized a nonuniform orthogonal mesh. The electrical field grid distance varies slowly in each direction, while the magnetic field resides in the middle of two adjacent electrical field. Though in this scenario the electrical field no long resides in the middle of two adjacent magnetic field points, which may result in first-order error locally, it has been shown by Monk1 that second-order error can still be achieved globally. In addition, we exploit the fact that within one Yee cell, the electrical field and magnetic field in each direction are half grids away to create an anisotropically filled Yee grid. This implementation maintains the accuracy of the cavity model with reduced grids and thus reduced cost of computation.

Figure 1.

Imaging cavity

Figure 2.

Modeling of a panel of bowtie antennas with anisotropically filled Yee grid