CUDA-based Local CT Reconstruction Acceleration And Metal Artifact Correction

While CT technology has been widely used in medicine and industry areas, people raise a higher requirement on the speed of its reconstruction. Therefore, developing a fast CT image reconstruction software is of great practical value. With the improvement of resolution of CT images, the data of CT reconstruction is becoming larger. If the object were still reconstructed according to the original vision, it would cause the increase in the calculation and the decline in reconstruction speed. Under normal circumstances, one is often interested in rebuilding the portion of a view, so the fine rebuild of local region provides a new thinking way to reconstruct CT image quickly, which could be achieved efficiently and conveniently. The CT image quality often suffers the effects of various artifacts during the reconstruction process. The metal artifacts caused by metal implants are common in CT artifacts, so it is of great value to study how to remove the metal artifacts quickly and effectively.This thesis first introduces the classic algorithm of FDK and CUDA acceleration and then introduces the design and implementation of the scheme of local regional reconstruction specifically; subsequently it gives a sketch of the operation process of the software and tests the software by the forward simulation data of the Shepp-Logan model and the data scanned by Micro-CT system. Finally, the problems of metal artifacts existed in reconstructed images are studied, which are caused by beam hardening effect. For the reconstruction of local region with high resolution and the elimination of the metal artifacts existed in reconstructed images, the main work of this paper can be summarized as follows:(1). The CUDA architecture is used to improve the CT reconstruction speed. In order to reconstruct the massive projection data, a scheme of automatic block is designed without changing the hardware configuration.(2). The software platform of the fine reconstruction in local region is designed and implemented to solve the problem that people often cares only about the fine information of the local region in a image view. First, an initial reconstruction in lowresolution is conducted for the whole view. Then the local region concerned is determined through the interactive select in three dimensional slices. Finally, a high resolution reconstruction of local region is conducted.(3). A normalized metal artifact reduction algorithm is adopted to reduce the metal artifacts, which will generate in CT image while x-rays pass through the object containing metal. In this thesis, the algorithm is designed and implemented for cone-beam CT, and the CUDA acceleration based on GPU is adopted for the processes that have a huge computation. Experimental results show that the method could remove metal artifacts in high efficiency.