Comparison with reconstruction algorithms in magnetic induction tomography.
Min Han,Xiaolin Cheng,Yuyan Xue
DOI: https://doi.org/10.1088/0967-3334/37/5/683
2016-01-01
Physiological Measurement
Abstract:Magnetic induction tomography (MIT) is a kind of imaging technology, which uses the principle of electromagnetic detection to measure the conductivity distribution. In this research, we make an effort to improve the quality of image reconstruction mainly via the image reconstruction of MIT analysis, including solving the forward problem and image reconstruction. With respect to the forward problem, the variational finite element method is adopted. We transform the solution of a nonlinear partial differential equation into linear equations by using field subdividing and the appropriate interpolation function so that the voltage data of the sensing coils can be calculated. With respect to the image reconstruction, a method of modifying the iterative Newton-Raphson (NR) algorithm is presented in order to improve the quality of the image. In the iterative NR, weighting matrix and L1-norm regularization are introduced to overcome the drawbacks of large estimation errors and poor stability of the reconstruction image. On the other hand, within the incomplete-data framework of the expectation maximization (EM) algorithm, the image reconstruction can be converted to the problem of EM through the likelihood function for improving the under-determined problem. In the EM, the missing-data is introduced and the measurement data and the sensitivity matrix are compensated to overcome the drawback that the number of the measurement voltage is far less than the number of the unknown. In addition to the two aspects above, image segmentation is also used to make the lesion more flexible and adaptive to the patients' real conditions, which provides a theoretical reference for the development of the application of the MIT technique in clinical applications. The results show that solving the forward problem with the variational finite element method can provide the measurement voltage data for image reconstruction, the improved iterative NR method and EM algorithm can enhance the image quality from different points, and the proposed image segmentation can make the lesions of the reconstruction image more likely to be identified.