Reconstruction Algorithm And Motion Artifact Correction Of Computed Tomography

Computed Tomography (CT) is a nondestructive and exact imaging technology which obtains image from X-ray or electron beam projection data by computer processing. With decades of development, CT has became one of the most advanced medical imaging technologies, and be widely used in clinical diagnosis and research.The most widely used CT reconstruction algorithm is Filtered Back Projection (FBP) algorithm. But FBP requires rigorous condition of projection data. Artifact appears in condition of the lack of projection views, the damage of projection data or too much noise. Moreover, to reduce dose is an urgent task, because of high dose rising carcinogenic risk. Unfortunately, inherent drawbacks of FBP block the further dose decrease. So researchers have been exploring more ideal reconstruction algorithms for incomplete project data. In recent years, compressed sensing (CS) theory which can break through the limit of sampling theorem to reconstruct signals using few of non-adaptive linear projection measurements was proposed. If CS theory can be applied to CT reconstruction algorithms, problems mentioned above may be solved smoothly.Artifact correction is an essential and hot area in CT research. Motion artifact which is caused by patient's conscious or unconscious motion is one of the typical artifact and destroys the consistency and completeness of projection data, The diversity and uncertainty of motion make it very difficult to estimate movement or reinstate projection data from the motion-damaged data, especially when human body contains metal implants, which will aggravate beam hardening and bring amount of noise. Motion artifact correction plays a crucial role in promoting the clinical application of CT.After mastering the CT imaging theory, we do an in-depth study on CS based CT reconstruction related technology and rigid motion artifact correction. In conclusion, the major works and contributions of this dissertation are organized as follows:1) CS based CT reconstruction algorithms and their applications in dose reduce and artifact correction is studied. After the successful implementation of the POCS-TVM algorithm that has the best overall performance in compared with other CS based reconstruction algarithm, we proposed an adaptive non-uniform sampling method based on the X-ray attenuation, which can reduce dose while ensure the acceptable quality of image. Besides, POCS-TVM based method to reduce the truncated data related artifact efficiently is proposed.2) Proposed a method to implement POCS-TVM algorithm fast. The main idea includes: sign effective areas of projection data and image, and then optimize the CT reconstruction model to get rid of unnecessary calculation; use ordered-iteration method to speed up the convergence of POCS; modify the procedure of TVM to avoid part of invalid calculation; apply CUDA based GPU to accelerate the whole algorithm.3) Proposed a fast and effective method to correct rigid motion artifact. First, transform projection data to frequency domain through projection slice theorem; second, estimate rotation from amplitude of frequency data based on the correlation in frequency domain; then estimate parameters of translation from projection data based on data consistency; finally, the artifact reduced image is reconstructed by introducing the estimated parameters. This method separated the estimation of translation and rotation,so that the whole rigid motion estimation process is simplified.4) Proposed a method to reduce artifact of rigid motion in condition of containing metal implants. First, use polynomial-correct method to regulate X-ray beam hardening and wavelet threshold de-noising method to reduce Poisson noise; second, estimate the rotation parameters via the correlation in frequency domain; and then estimate translation via the characteristic points'geometric in Radon transform; finally, compensate the motion effect with the estimated results and correct metal artifact by linear attenuation method. This dissertation supported by the National Natural Science Foundation of China(60771007)& Innovation Foundation for Postgraduates of USTC.