CT Image Reconstruction Method for Local Imaging of Arbitrarily Shaped Convex Region with Sparse Sampling Scheme

Abstract

The increasing utilization of x-ray computed tomography (CT) has generated a great concern over the potential cancer risk associated with hazardous radiation dose. Considering that the effective dose from a CT scan on average is about 10mSv while that of background radiation is about 2.4mSv per year, reducing the radiation dose in CT scan is of great importance for patient's health, particularly for those who have undergone repeated CT scans. For this reason, recent technical developments have focused on lowdose imaging without compromising the image quality. Recently, most major CT manufacturers such as Siemens Medical Solutions, GE Healthcare, and Philips Healthcare have developed iterative reconstruction methods capable of reducing radiation dose by 30-60% without increasing noise. However, dose reduction cannot be emphasized enough, as long as it can be reduced without degrading the image quality. In the recent decade, fundamental advances have been made in tomographic image reconstruction by introducing prior knowledge, sparseness or compressibility. Compressed sensing (also known as compressive sampling or sparse sampling) is an innovative signal processing technique to find solutions of underdetermined linear systems. In classic sampling theory, most of the acquired data at the Shannon-Nyquist rate can be discarded during the data compression stage without much loss if the data is sparse or compressible. On the contrary, compressed sensing taking advantage of the sparseness or the compressibility allows the entire information to be recovered from a relatively few measurements, which means low radiation dose in CT scans. In CT scans, one straightforward way to reduce the radiation dose is sparse sampling in an angular direction called sparse-view CT. Another is to minimize the area to irradiate x-ray by limiting the field-of-view of illumination only to the region-of-interest (ROI), which is called ROI-CT. However, both of them produce contrast anomalies such as streak artifacts and band-bright artifacts in classic image reconstruction. Various methods have been investigated to remedy these contrast anomalies and recently introduced two methods have improved the image quality using the popular compressed sensing techniques. Both of them use fully sampled interior projection data and one of them does a handful of global projection data additionally. However, sparse sampling in the interior projection data would be better as in the sparse-view CT. In the fields of image compression, the ROI functionality is of great importance for preserving the ROI inside with a high quality but outside the ROI with a low one. In most cases of image compression, the ROI is set to a circular shape though the actual ROI is an arbitrarily shaped ROI. In terms of image compression efficiency, the arbitrarily shaped ROI that minimizes the ROI size by encompassing only the actual ROI is preferable for the purpose of reducing bits in image compression. Furthermore, the arbitrarily shaped ROI is highly desirable because it can reduce the dose in CT scans as it reduces bits in image compression. In this dissertation, in order to reduce the radiation dose while preserving the image quality, an image reconstruction method is proposed for local imaging of an arbitrarily shaped convex region with sparse sampling scheme. More specifically, a sparse sampling scheme in the interior projection data and an arbitrarily shaped ROI that minimizes the ROI size has been employed. The proposed image reconstruction method is a sort of multi-resolution acquisition technique and uses two sets of projection data, interior projection data with a relatively high sampling rate in the angular direction, and sparseview exterior projection data with a much lower sampling rate. The proposed image reconstruction method called the exterior view assisted interior tomography (EV-IT) solves a constrained total variation (TV) minimization problem for the interior projection data, which is assisted by the exterior projection data in the CS framework. Then, a new collimation method is proposed to make a non-circular convex-shaped ROI. To make an ROI with an arbitrary convex shape, dynamic collimations are necessary to control the field-of-view at each view angle. Finally, the proposed method reconstructs images by connecting the dynamic collimation and EV-IT. In numerical simulations and experiments, it has been proved to effectively suppress bright-band artifacts and streak artifacts arising from sparse sampling and data truncation. Thus, the proposed methods are expected to contribute to reducing the x-ray dose significantly in CT scans if the dynamic collimation for the arbitrarily shaped convex ROI is realized in real CT machines.