Atlas-Based interior tomography

Interior tomography is image reconstruction of a region-of-interest (ROI) located inside a scanned object. It has major effect in reducing the radiation dose. It was believed previously that the solution of interior problem was not unique. In the last decade, it has been learned that the interior problem can be solvable if sub-region inside the ROI is recognized or if the ROI is piecewise constant. Eventhough, such cases are not always available and the existence of recognized/currently known sub-region inside the ROI is not confirmed in many situations. In this paper, we introduce the use of prior knowledge from probabilistic atlas in solving the interior problem. The atlas is constructed by using a set of computed tomography (CT) images which are scanned in advance for different patients and fitted by Gaussian mixture model (GMM). Then, expectation maximization (EM) algorithm is used to estimate the mixture parameters. Whereas the probabilistic atlas is generated from advance scan of various patients, the prior information is always available and confirmed for all applications. We develop a statistical reconstruction method where the cost function is designed to incorporate the information of the probabilistic atlas. Both experimental results of the simulation and real data indicate that the proposed method remarkably improve the image quality and minimize the effect of DC-shift. Surprisingly, we observe a high quality reconstruction concerning the region outside the ROI.