Experimental electrical impedance tomography (II)

Abstract

The report reviews the essence of Electrical Impedance Tomography (EIT) and studies the different variations of reconstruction algorithms to evaluate their sensitivity and potential for medical imaging. EIT is an interesting method for clinically monitoring patients due to its non-invasive methodology in providing continuous imaging of body part, organ, or object impedance. However, the image reconstruction in EIT is not sequential and an incomparable inverse problem. This is due to the use of Tikhonov prior with L2 regularization, a common way to solve EIT problems which that it always smooths out sharp changes or cease regions of the reconstruction. [1] Hence the use of L1 regularization in image reconstruction enables to direct this difficulty. L1 norm on data terms provide estimation strongly built to outliers. In view of this, L1 norm on regularization terms reconstructs sharp spatial profiles. The report comprises of a comprehensive study made on the performance of two different groups of regularization algorithm namely L1 and L2 regularization when challenged by Gaussian noise. L1 regularization group consists of least absolute shrinkage and selection operator (LASSO), Elastic Net and Total Variation (TV) algorithms. While L2 regularization group consists of Newton's One‐Step Error Reconstructor (NOSER) prior, Tikhonov prior and Laplace prior algorithms. Simulated results of L1 and L2 regularization methods via MATLAB were compared to determine which group of regularization can produced improved quality of image reconstruction yet withstand higher noise levels in measured voltages.