A new image reconstruction method for electrical impedance tomography

Electrical impedance tomography(EIT) is a functional imaging technique, which has potential application prospect in clinical diagnosis. It is well known that image reconstruction in EIT is a highly ill-posed, non-linear inverse problem. So far, various reconstruction methods have been used in EIT, among which the Newton-Raphson method is regarded as the most effective one, but it suffers the mathematical difficulty and the low resolution of the reconstruction which is far from the requirement of clinical application. In this paper, a quite different image reconstruction method for EIT is presented to solve the above problems. In the new method, a neural network such as BP is used to solve the non-linear inverse problem between the impedance variations inside body and the voltage changes measured at its surface with no need of computation of potential fields. The training sets of neural network are chosen from the solution of the forward problem of EIT in which finite element method (FEM) is used. After the non-linear relation function has been decided, the static image reconstruction can be accomplished by iteratively solving the forward problem with FEM until the voltage difference between measurement and calculation or impedance change is small enough. The provided method avoids calculating Jacobian matrix and solving ill-conditioned equations. Furthermore the resolution of the reconstructed images based on the new method is much more higher than other methods with the same numbers of electrode and electrical current pattern. The computer simulation results demonstrate that the new reconstruction algorithm can be converged very quickly with a priori knowledge.