Estimation and Use of Prior Information in FEM-CSI for Biomedical Microwave Tomography

Abstract

Prior information is used to improve imaging results obtained using the finite-element contrast source inversion (FEM-CSI) of a microwave tomography (MWT) dataset collected as part of a forearm imaging study. The data consist of field measurements taken inside a prototype MWT system that uses simple dipole antennas and a saltwater matching medium. Initial images of the 2-D cross-sectional dielectric profile of the individuals' arms are reconstructed using FEM-CSI. These initial "blind" imaging results show that the image quality is dependent on the thickness of the arm's peripheral adipose tissue layer: Thicker layers of adipose tissue lead to poorer overall image quality. The poor image quality for arms with high levels of adipose tissue is not improved by changing the matching fluid's complex dielectric constant. Introducing prior information into the FEM-CSI algorithm in the form of an inhomogeneous background consisting of an adipose layer surrounding a muscle region provides substantial improvement of the image quality: The internal anatomical features of the arm are resolved for each of the five datasets. Two methods are employed to estimate the arm periphery and adipose layer thickness from the blind imaging results: manual estimation and a novel image segmentation algorithm based on global optimization using simulated annealing.Index Terms-Contrast source inversion, inverse scattering, microwave imaging, prior information, simulated annealing.

Breast Imaging Using Microwave Tomography With Radar-Based Tissue-Regions Estimation

“…1(a), as it allows one to incorporate prior information in the form of an inhomogeneous background containing known information about the dielectric properties and location of anatomical regions in the breast ( Fig. 1(b)) [13]. In addition, the algorithm offers the benefit of performing inversion on triangular meshes, which can accommodate the irregular shapes of anatomical structures [5].…”

“…For example, in [13] it was shown that using information about the fat thickness as an inhomogeneous background significantly improves the quality of the reconstructed images of human forearms [11,13].…”

Full-Vectorial Parallel Finite-Element Contrast Source Inversion Method

“…As the bulk of the memory used in the MR-FEM-CSI algorithm is required for storing the LU-decomposition of the FEM operator, memory limits are not bound by the size of the FEM mesh itself thus the mesh can be duplicated in its entirety on each processor. Nevertheless, if one wished to use a fully distributed mesh representation, additional "ghost" elements are required when applying MR for evaluating the surface integral in (9). In our implementation, the mesh is distributed amongst the processors.…”

“…Traditionally 3D CSI has been based on integral equation formulations of Maxwell's equations on a regular grid [7] and, similar to MR-GNI, requires determining the appropriate Green's function. An FEM discretization of the problem enables arbitrary boundary conditions [8], inhomogeneous background media [9] and variable mesh density [10] to be built into the model directly. Recently, these benefits have been shown for the 2D scalar [8,9] and vector FEM-CSI [11,12] using synthetically generated as well as experimental data.…”

The Time-Harmonic Discontinuous Galerkin Method as a Robust Forward Solver for Microwave Imaging Applications

“…Our desire to model both dielectric and magnetic materials and sources is motivated by biomedical microwave imaging (MWI) applications. PDE formulations of Maxwell's equations readily support inhomogeneous background materials and boundary conditions, and have been recently exploited in MWI algorithms by using an FEM forward solver [17,18]. Recent research suggests the use of magnetic contrast agents for cancer-detecting MWI [19] and so we consider forward solvers that support both magnetic and dielectric materials.…”