Study On Metal Artifact Reduction In Computed Tomography

CT has become an indispensable tool for clinical diagnosis in hospital and research method since the first clinical CT equipment had been published from1972. Not only can be used for diagnosis of human disease, CT image also be committed to many other applications. However, difference always exits between actual and ideal CT system that reflected in the inaccuracy of data. Different kinds of artifacts will be produced in the reconstruct image with these inaccurate data. According to the difference of reasons and characteristics of artifacts, there are geometric artifacts, hardening artifacts, scattering artifacts, motion artifacts and metal artifacts. The metal artifacts play an important part in the quality of CT image.It is inevitable that patients carry or contain metal when scanned with CT in medicine. The metal produces bright and dark streak artifacts in image after reconstruction. Because the pathological features of human tissue and the performance of artifacts are different, the artifacts will not lead to misdiagnosis in normal conditions. However, when the intensity and area of artifact is large enough, the artifacts reduce the image quality and even make the image not be read.The causes of metal artifacts are rather complex. The metal artifacts can lead to beam hardening, partial volume or the dynamic range of electronic devices working in data acquisition. In addition, there is evidence that the motion of metal is one of the main reasons of artifacts. Obviously, lots of reasons can cause metal artifact. But the basic reason is associated with the metal character of high attenuation. The high attenuation makes the X-ray hard and also aggravate scattering.Aiming at this problem, scholars have put forward various metal artifact reduction (MAR) methods. These correction methods can be divided into three categories:iterative reconstruction method, projection interpolation method and the mixed method.Iterative method is also called "step by step approximation method". It is a method commonly used in solving the matrix equation. Iterative method assumes that the image is uniform and sets any value for the matrix at the beginning, the theoretical values are compared with the measured projection values and then correct the difference between them. Repeat the steps until the assumed values and the measurement values are same or in the acceptable error range. The iterative algorithm can be divided into two categories:the algebraic iterative reconstruction method and the statistical iterative reconstruction method.In1970, Gordon et al first used the concept of algebraic reconstruction technique (ART) in the field of image reconstruction. ART algorithm is an iterate process that correct image vector. The reconstruction area is discretized into a digital image and being projected from different angles, linear equations are established by the projection data. The attenuation coefficient distribution of the reconstruction area is acquired through iteration of the equations. The attenuation coefficient is the pixel value of the image here that can be regarded as unknowns of linear equations. When the number of image pixels is certain and the number of equations established is enough, the attenuation coefficient values can be determined by solving the linear equations. Then the image is reconstructed. When the CT image contains metal artifacts, the absence of projection data can be considered as the missing of linear equations. Only the equation number is large enough the image can still be reconstructed.Statistical iterative reconstruction method is composed of the objective function and iterative method. Common objective functions are maximum likelihood (ML) function, maximum a posteriori (MAP) and minimum mean square error. Iterative methods are expectation maximization (EM), the steepest descent method and the conjugate iterative method. Different objective functions and iterative methods form different statistical iterative reconstruction method, such as maximum likelihood expectation maximization (MLEM) and maximum a posteriori expectation maximization (MAPEM).Iterative method uses iterative reconstruction algorithm to remove metal artifacts. It can remove metal artifacts effectively and restrain noise. But the quite large computation and slow speed make the method difficult to be applied.The projection interpolation method is generally based on filtered back projection (FBP) algorithm. Compared with iterative method, the FBP algorithm has less calculation, high speed and is more practical. However, it is very sensitive to high attenuation metals, significant artifacts appear in the reconstructed image when data change suddenly. The sudden change of projection data in metal track edge causes artifacts around the metal object in the reconstructed image when the scanned object contains metal. The projection interpolation method is based on the idea that the metal artifacts can be removed if the sudden change of projection data which adjacent metal can be avoided. The process is as follows:firstly, the original image is reconstructed by FBP algorithm with the initial projection and the metal is separated from uncorrected image which contains metal artifacts. Secondly, the location of metal trace is determined with forward projection of the separated metal. Thirdly, interpolate the metal region using the projection of non-metallic area which surrounding the metal and reconstruct the interpolated projection. Finally the corrected image is acquired by adding metal to the reconstructed image. Metal division and projection interpolation are two key steps of interpolation method that both of them play an important role in the finally corrected output.The commonly used interpolation methods are linear interpolation, polynomial interpolation and spline interpolation. In1987, Kalender et al proposed linear interpolation. Linear interpolation is the most simple interpolation method and can eliminate the streak artifacts which produced by metal to a large extent. But the left and right end points of interpolation interval are not smooth, which produce peaks in the two end points during reconstruction. The corrected image produces new streak artifact in the metal edge. Spline function interpolation can get a smooth curve and avoid strip artifacts which are caused by mutation of boundary projections. But cometary bright streaks appear in the reconstruction image with this interpolation method. The overall correction effect of this method is not equal to linear interpolation. All the above interpolation methods can remove metal artifacts in a certain extent, but the metal information in the reconstruction results will be lost.The projection interpolation method is the most popular algorithm of metal artifact reduction because it has simple theory, fast speed and easy to realize. But it only has significant effect when deal with simple metal objects.It is easy to think that correct metal artifacts by integrating the advantages of iterative method and interpolation method due to their different advantages and disadvantages. In2005, Xia D et al proposed a typical mixed method by combining FBP and EM. The metal region is reconstructed using EM iterative after the metal region is identified, followed by interpolation of the metal region and then reconstruct the entire projection with FBP algorithm. The final result is the composition of metal area and non-metal area. The mixed method is still based on the frame of interpolation method. With iterative method the algorithm performs better result for the reconstruction of metal area, while the reconstruction region around the metal is same with the interpolation method.The mixed method is combined with interpolation method and iterative method, which has advantages of the two algorithms. It is faster and can reflect the structure of metal objects but causes distortion in the area around the metal. There is still great limitation in practical application.The present metal artifact correction methods all use original data of CT images which obtained from the detector directly, or use the original date in some way to replace the FBP data. While in many cases the result can only be acquired from contaminated image as the raw data cannot be obtained. In order to overcome this problem this article proposes an algorithm based on Hough transform and mathematical morphology. The input value of the algorithm is CT image. The artifacts are eliminated with morphological filter after the conversion from Cartesian coordinates to polar coordinates. Firstly, extract the streak artifacts trough the line detection and Hough transformation, and find the coordinates of metal center trough the curve fitting and least square method. Secondly, transform the initial image containing metal artifacts to polar coordinates in the point of view of the metal center coordinates. Thirdly, select appropriate structure elements and morphological filter to remove artifacts in the polar coordinates image. Finally, transform the filtered image from polar coordinates to Cartesian coordinates. The algorithm not only considers the brightness of the artifact pixels, but also regards the geometric characteristics of the artifacts and compares the effects of different morphological filters and structure elements of different size. The experiment results show that this method can effectively reduce streak artifacts with quite slight blur in local area. The algorithm can be widely applied or as a pre-processing step for other artifact correction method because of the short computation time. The algorithm is less effective for image without serious pollution due to the quality of FBP image and the morphological filter. Research in the future will focus on the adaptability of the morphological filter.A major disadvantage of simple interpolation method is the loss of information, especially the edge information of metal area, which will cause blurring of the corresponding edge in the image. Considering this problem, this article proposes a normalized metal artifact reduction (NMAR) algorithm. The algorithm still adopts the frame of interpolation method. Firstly the image is filtered with anisotropic Gauss filter to remove noise and smooth the streak artifact, followed by K-means clustering algorithm to produce metal image and prior image, and project the related images to get corresponding sonograms. Then using the forward projection of the prior image to normalize and interpolate projection and multiplied by the projection of the prior image to inverse normalize. The final image is reconstructed with corrected projection. Experiments show that NMAR algorithm performs good correction results of metal implants. Even for artifacts close to metal of high intensity the algorithm can eliminate the most of them. After the NMAR algorithm there is almost no artifact away from the metal implants. Compared with the simple linear interpolation method, the proposed algorithm performs better result and avoids the missing of information in the metal area and production of new artifacts.At the end of this paper we summarize the work briefly and prospect future work.