| Computed tomography (computed tomography, CT) is one of the most advanced medical imaging technology in modern times. With the advantages of non-contact, non-destructive, high resolution and non-overlapping image, CT becomes an indispensable important technique in various fields such as clinical medicine, industry, material and biology. CT imaging technique is the use of X-ray projection at different angles to measure for all information of tomograms within the scanning range. Compared with the traditional spirial CT, Cone beam CT (CBCT) performs better with the advantages of fast scanning speed, high spatial resolution and high radiation efficiency, and it can perform three-dimensional imaging without damage to the human body, which become a new hotspot and direction in the field of medical imaging diagnosis.In the CBCT system, FDK algorithm is usually used for image reconstruction. But FDK algorithm is strict with the geometric model of CBCT system. The requirement to the geometric model has the following two points:First, the line connected the X-ray source and center of the detector must cross the rotation axis, and it projects perpendicular to the detector plane. Second, the axis of rotation must be parallel to the column direction of the detector. After manual installation of the CBCT hardware system, the whole system can’t strictly satisfy the two requirements with the geometry model of FDK reconstruction algorithm due to the limitation of mechanical accuracy. Therefore, there exists a certain geometric deviation between actual geometric model of CBCT system and ideal geometric model with the requirements of FDK algorithm. The reconstruction artifact caused by the geometric deviation, which called geometric artifacts. Geometric artifact reduces the quality of reconstructed images, and influences the doctor to diagnose disease, which reduce the medical quality and safety. As a result, the calibration of geometry artifacts has important significance to obtain high quality reconstruction image for CBCT system.CBCT geometric calibration algorithm is mainly divided into analytical algorithm and iterative algorithm. Analytical method is always based on the premise of certain ideal conditions and corrects only part of the system parameters, which reduces the complexity of the problem but narrows the application range of the geometric correction. Iterative method used projection information or reconstruction image quality standards as constraints to solve the geometric parameters of the system by optimization algorithm.The traditional analytic methods of geometry correction usually need to make an accurate calibration phantom and measure the information of the mark point on the phantom as the known conditions. The projection data is achieved by multi-angle scan with the calibration phantom. With the projection data, the geometric parameters of CBCT system can be extracted. Owing to short calibration time, simple manufacture of calibration phantom and strong applicability, a large part of the research achievements in analytic geometry correction algorithm used as proprietary technology or commercial secrets. Therefore, it becomes the mainstream of modern geometric correction for CBCT system. Through the analysis of the principle of geometric correction method, the development trend of the quantity of projection information for the calibration phantom transform from the multiple perspectives to the single view gradually. For the simple structure phantom, it requires scan the phantom in a wide angle scope to collect enough projection data to satisfy the conditions to solve the equations. But for the complicated structure phantom, the problem can be work out with small angle scope or a single view.Along with the progress of computer technology, iterative algorithm has been greatly developed. It overcome the shortage of demanding a calibration phantom in traditional methods and relies on the projection data of the reconstructed object as the self-calibration dataset of geometric parameters.Currently iterative algorithm can be divided into the following two categories:First, the geometric parameters can be directly solved based on the characteristics of the projection images of the detected object. The second type is based on the feature of geometric artifact of the reconstructed images to construct the objective function to quantify geometric artifacts. It finds optimal geometric parameters corresponding to the minimum geometric artifacts through the optimization algorithm to achieve the purpose of calibration. Although iterative algorithm calibrates the artifact with high precision, but it always suffers some problems such as difficult initial value selection, local optimum and slow speed of calibration. So this type of algorithm is not the mainstream of the CBCT system calibration algorithm.For iterative algorithms, it often requires to reconstruct a part of CT image to quantify assessment of the geometry artifacts. Therefore, the speed of reconstruction is a concern. With the progress of the development of computer software and hardware, the computing industry convert from "Central Processing" with the central processor CPU (Central Processing Unit) to the "cooperative management" of CPU and GPU (Graphics Processing Unit). NVDIA introduced a GPU product based on CUD A (Compute Unified Device Architecture). CUD A has powerful parallel computing and huge storage to support its high-speed parallel computing. Its working principle is:The host sends commands to the Graphics card and GPUs compute in parallel to process tasks in the Graphics. Finally the computed results return to the host. Above all, the working process can save the calculation time greatly.This article mainly works on the research of the geometric calibration for cone beam CT systems. It first designs and manufactures the calibration phantom specially for the calibration methods. Through the projection information and a priori knowledge of calibration phantom, geometric errors in the cone beam CT system can be calculated used for correction for the reconstruction algorithm and the quality of CBCT images is improved.This paper introduces the significance and research status of geometric correction. Afterwards the imaging principle of cone beam CT system and FDK cone beam reconstruction algorithm are introduced in this paper. And the cone beam CT imaging system geometric model is set up, and then this article focuses on geometric artifact correction technology of the cone beam CT system. It mainly improved and implemented the following three methods:1. In view of the existing cone beam CT equipment, a kind of circular orbit geometric correction calibration method based on image sharpness has been realized and improved. The algorithm is to build a optimization model based on the sharpness of images as the objective function. It uses the existing analytical method to calculate geometric parameters as the initial value to narrow the search scope. In each iteration, the algorithm uses the current geometric parameters to correct the artifacts of the reconstructed images. And then the quantitative assessment of the specific layer of the reconstruction images, the sharpness value is used as objective function. Change the geometric parameters and then enter the next iteration until it meets the terminating condition. Since the optimal geometric parameters of the solution will appear near the analytical initial value, one dimensional linear search can solve the optimal solution of geometric parameters to maximize the sharpness. Compared with the existing analytical method, the experimental results show that this method has high precision and the advantages of low computational complexity and it can effectively eliminate the inherent error in analytic method. The geometric artifacts of the reconstructed images are significantly calibrated. Aiming at the problem of slow reconstruction speed in the process of calibration, the method uses the graphics processor GPU for 3D reconstruction to accelerate.2. Draw lessons from the pinhole camera model, this paper proposes a geometric correction method of cone beam CT in circular orbit. Through scanning a specific calibration phantom within 360 degrees scope, the ellipse parameters are extracted in the imaging area. Circle point and polar constraint conditions in the higher geometry are used to calculate the intrinsic matrix of cone beam CT system. Based on the obtained parameters, the extrinsic parameters of cone beam CT system can be achieved by the geometric method and ellipse parameters. Compared with other geometric correction method, this method can calculate all of the geometric parameters of cone beam CT system and it establishes a complete mathematical model to describe the existing cone beam CT system. The calibration phantom can be made simply and this method acts with strong applicability. Experimental results show that the calibration precision of internal parameters and external parameters calculated by this method for cone beam CT geometric is 0.193% and 0.2%. Finally 3D reconstruction of the simulated head phantom is performed in the calibrated cone beam CT system. Comparison between misaligned and aligned CT images is used to assess the effect of geometry calibration method.3. The geometry of cone beam CT system is regarded as the pinhole camera model. A DLT (Direct Linear Transformation) geometric correction method based on mapping matrix is implemented. In this paper, a cylindrical calibration phantom with spiral distributed markers is designed and constructed. The three dimensional space coordinates of the center of mass of the marker is used as a known condition. The projection data of the calibration phantom within 360 degrees is acquired and the two dimensional coordinate is extracted from the projection of the center mass of markers. An over-determined equation about mapping matrix can be solved by using the dataset with three-dimensional coordinate corresponding to two-dimensional coordinate. And then the mapping matrix corrects the reconstruction algorithm at every angle to obtain the accurate CT images. Through processing an experiment with the projection of Shepp-Logan digital phantom and a simulated head phantom, the feasibility of this method is verified. Experimental results show that the method can effectively remove the geometric artifacts in the reconstructed images caused by geometric deviation by comparing the misaligned and aligned CT images. This method belongs to the category of geometric calibration at single angle. Compared with the traditional geometric correction method under the multiple perspectives, its advantages are the following two points:1. It is applicable in many different trajectories of CT system, such as detector and X-ray source rotate synchronously around the rotation axis or X-ray source rotates around rotation axis within a certain range when detector stays static, etc.2. If the mechanical jitter or position deviation appears in the process of spinning, for traditional geometric correction method under the multiple perspectives, it will increase the computational complexity of geometric correction algorithm and reduce the calibration precision. But the algorithm in this article can correct the reconstruction algorithm under the single view and it can effectively eliminate the impact to the CT images caused by the mechanical instability in the process of spinning, which has the ability to provide high quality of reconstructed images.Precise geometric calibration method is an access to achieve high quality CT image in the cone beam CT system. After artificial installation of CBCT hardware, it necessary to do the geometric calibration to eliminate the geometric artifacts on images due to mechanical error in CBCT system. The geometric calibration methods of cone beam CT system in this paper makes a preliminary exploration and obtained some preliminary research results, but it still needs further research to improve the shortcomings of existing methods. |
PDF Full Text Request