Research On CT Image Metal Artifact Reduction Using Optimal Hybrid Methods

Computed Tomography (CT) is a medical imaging method employing tomography. Despite recent progress in CT, artifacts still significantly limit several important clinical applications. It can seriously degrade the quality of X-ray CT images, sometimes to the point of making them diagnostically unusable. An Artifact is any distortion or error in the image that is unrelated to the subject being studied. For X-ray CT, artifacts are any discrepancy between the X-ray CT numbers represented in the image and the expected X-ray CT number based on the linear attenuation coefficient.Among the different complicated image artifacts, the metal artifact has been more pronounced. It degrades image quality as well as hides areas of pathology, so it is important for us to know the appearances and causes of artifact. The causes of metal artifacts are quite complicated. Depending on the shape and the density of the metal objects, the artifact appearance can vary significantly. When the cause of the metal artifact is dominated by beam hardening, it can be corrected by adequate beam filtration. When the metal objects are highly attenuating, under range in the data acquisition electronics often results. When this occurs, image artifacts similar to photon starvation appear. This type of artifact is dominated by high-intensity streaks. When combined with the beam-hardening artifacts caused by metal, both shading and streaking artifacts appear in the reconstructed image. A combined beam-hardening correction and adaptive noise filtering approach should produce a further improved image. Partial volume is another cause of metal artifact, partial volume occurs when an object is partially intuded into the scanning plane. As the slice thickness increases, the likelihood of partial volume occurrence increases.No matter what kinds of causes of metal artifact, it can appear as geometrical inconsistencies, blurring, streaks or inaccurate X-ray CT numbers. And streak artifacts are the most common distortions or errors that affect the quality of X-ray CT images. A streak is usually caused by inconsistency present in isolated measurements, the inconsistency could be the result of an inherent problem associated with the data collection process or abrupt changes between views, when inconsistency occurs in the projection data set, the reconstruction process is no longer able to properly combine the positive and negative contributions, and lines or streaks will result. All these artifacts significantly degrade X-ray CT image quality and limit the usefulness of CT for many clinical applications because tissues in the plane of the metal appliance are severely obscured. Hence, there is a high demand on X-ray CT metal-artifact reduction for sound diagnostic investigation as well as for an accurate planning in image-guided surgery.Methods for MAR aim at the improvement of the quality of images that are affected by metal artifacts. The convertional MAR procedure consists of the following steps: reconstruction of a preliminary image, segmentation of metal objects within this image, reprojection of the segmented objects, linear interpolation of the obtained traces in the sinogram and finally reconstruction of the corrected raw data. Several corrective methods have been studied in the past to reduce streak artifacts caused by high attenuation material. These approaches can be classified into iterative reconstruction methods, projection interpolation, and filtering methods, or combinations of those. Based on the previous discussion, the interpolation and the reconstruction algorithms have their respective advantages, so we combine these two methods with the effective segmentation. In this dissertation, we investigate the interpolation and reconstruction methods, and put emphasis on the hybrid interpolation and portion reconstruction. The study covers the image filter, segmentation, interpolation and reconstruction methods which aim to improve the MAR accuracy.The main contributions in this PhD dissertation are as following:First, Using the acquired raw projection data, the Adaptive Anisotropic Gaussian Filter reduces the noise content and to smooth streak artifacts in the image with metal artifacts. And it shows good results for the removal of noise artifacts, especially at greater distance from metal objects.Second, segmentation method named Mutual Information Maximized Segmentation (MMS) is proposed to segment the filtered image. Most researchers have adopted simple threshold methods to segment the metallic objects, the threshold-based methods may produce inaccurate segmentation of the metallic objects and, hence, the information from structures surrounding the metal may be lost. To improve the accuracy of MAR, we use the MMS technique, in MMS method, simulated annealing is used for finding the global minimum. Experiments show that the segmentation by MMS has the more entropy value, which is equal to MI between the original image and its segmentation, than that by mean-shift and FCM.Third, A hybrid interpolation method to improve the MAR result. Different linear and polynomial interpolation techniques have been developed for estimating the "missing" projection data. Linear interpolation is the most commonly used interpolation methods. With the traditional interpolation algorithm itself, it fails to accurately outline the borders of metal objects, and can not restore a good metallic appearance. This attributes faults to both a single interpolation method and accurate segmentation algorithm:Using linear interpolation for MAR to remove strip is very effective, but it depict the immediate vicinity of metal objects, since these image parts are strongly influenced or even erased by the interpolation process. The method nevertheless gives good results in the near to intermediate area by correcting CT values that have been influenced by scatter and beam hardening effects. At greater distance the method tends to introduce new artifacts, especially along lines connecting the metal objects and objects of high density (e.g., bone) or edges of the patient or the table. This is due to the interruption of the traces of such structures in the sinogram by the interpolated metal trace. Polynomial spline classes could give detail information on metal internal reflection based on its interpolation characteristic determine the smoothness after interpolation, at the same time, it is not particularly suitable for artifacts to effectively clear the stripes. So a single interpolation method can not completely meet the clinical requirements. Based on this, a merging method is proposed to use interpolate, that is, a polynomial interpolation is selected to interpolate the metal part, and stripe metal interpolation is with radiation artifacts. Results show that the merging interpolation improves the images over the usage of only one of the interpolation methods.At last, we present a novel method to reconstruction interpolated projections. Although the mechanism is clear, conventional FBP methods are computationally efficient but produce image artifacts when complete and precise projection data are unavailable. it is not an appropriate procedure to cope with the inconsistencies in the Radon space. To suppress the metal artifacts, iterative reconstruction methods have been successfully applied that avoid the corrupted data. However, this approach is computationally expensive and not practical for clinical imaging. With this hybrid reconstruction method, first, the acquired raw projection data is used to separate contributions from the metal and non-metal objects. Next, reconstructions of metal and non-metal images are separately obtained from their estimated projections. A final image is formed by appropriately combining the individual images. With this method, images reconstructed contain virtually no artifacts. For metal objects with complex structures, the hybrid algorithm can also yield images containing fewer severe streak artifacts than those reconstructed using the FBP algorithm alone. The separation of the metal and non-metal projections allows for the effective reduction of metal artifacts, while the use of different algorithms offers efficient and effective reconstruction of the individual images. The results of these studies show that the hybrid approach can effectively reduce the dark regions and streak artifacts in images caused by the metal object. Therefore, the hybrid approach provides an effective correction for the metal artifacts in CT images.In the future work, we want to explore the application of special goal, such as the number of test images will be increased. Secondly, we think that hybrid interpolation can be looked as a generic model and try to use it in other scope beyond MAR. Thirdly, we will consummate the local iterative reconstruction method.