Parametric Detection and Classification of Compact Conductivity Contrasts With Electrical Impedance Tomography

Abstract

Electrical impedance tomography is a noninvasive and cost-effective imaging method that is increasingly attractive in the field of medical diagnostics. Several health conditions, such as stroke and solid tumors, are characterized by compact conductivity anomalies surrounded by a fairly regular background. Commonly employed voxel-by-voxel reconstruction methods for impedance imaging share the disadvantages of high computational cost and substantial sensitivity to measurement noise and imperfections in the electrical model describing the domain of interest. We present a special purpose algorithm for automatic detection and identification of compact conductivity variations. The technique exploits a priori structural information and, by reconstructing only the limited number of parameters required to describe a compact conductivity contrast, does not depend on a critical regularization parameter. The most demanding kernels are implemented to run on graphics processing units to accelerate computation. The parametric reconstruction is quicker and more robust than widely employed approaches with respect to measurement noise and imperfections in the electrical model, as shown by computational analysis performed on a segmented head domain and experimental measurements acquired on a cylindrical phantom. When the goal is quick detection of compact conductivity contrasts in complex 3-D domains, the inclusion of specific constraints relating to the problem considered leads to enhanced quality of reconstruction, making the presented technique a promising alternative to common voxel-by-voxel reconstruction methods.