Research On Image Registration Technology And Reconstruction Technology In IGRT

The rapid development of computer technology and electronic imaging technology, especially the wide application of cone-beam CT on linear accelerator, greatly improves the accuracy of image-guided radiation therapy and make the tumor treatment into an era of precision radiotherapy. But there are several factors that affect radiation precision: the displacement of tumor due to respiration or adjacent organs motion, the change of tumor volume and shape in the procedure of treatment, the change of patient body geometric location. These reasons result in the deviation of target position at the time of treatment from those approved in the treatment planning. This position deviation can cause the dose distribution deviation from the treatment planning, and can cause a vital organ adjacent to tumor radiated with high dose while the tumor sparing from the radiation. This can increase the probability of occurrence of patient complications. How to improve the accuracy of patient positioning and the precision of X-ray irradiation dose is a key research in the field of medical image processing. This research is very important to improve the patient rehabilitation rate.In order to improve the patient positioning accuracy and reduce the safety margins, we dedicated ourselves to the patient setup of CBCT image-guided radiotherapy system. In addition we discussed the coincidence between tumor target centroid and the origin of the coordinate system in detail. The coincidence problem is the key step in correcting the patient setup.In comparison to B spline, NURBS can use non-uniform grid in registration so it has a higher accuracy and higher speed. So we propose to use NURBS to register the planning CT image and the CBCT image. In addition, we propose a new contour-based fast similarity metric algorithm. Before introducing the fast similarity metric algorithm, we discussed some of the common segmentation algorithms in medical image segmentation fields, such as the Levelset-based segmentation algorithm, Contourlet based segmentation algorithm, the segmentation algorithm based on Markov random field, neural network-based segmentation algorithm, automatic image segmentation based on Bayesian method, active contour model-based segmentation method etc. These methods can be classfied into the region-based segmentation methods, the edge detection based segmentation methods, fuzzy set theory based methods, neural network-based method, Atlas-based segmentation methods etc. In our experiment, we use the Snake model to segment image contour and the experimental result is satisfactory.In this dissertation we studied the dose distribution of CBCT image and planning CT image. The dose distribution is an accurate quantitative analysis parameter. Optical flow field-based registration method can find the subtle changes between the time series of images. We want to use the optical flow field model to register more accurately the CBCT image and the planning CT image. This method will be helpful to get an accurate dose distribution. The registration among the time series images belongs to the monomodal alignment while the CBCT image and planning CT image registration belongs to the multi-modal registration. So we cannot use directly the optical flow field model to register the CBCT image and planning CT image. We presented a new registration scheme:first the rigid registration and the deformable NURBS-based registration are executed; then execute the alignment based on the optical flow field. In order to resolve the inconsistencies of pixel gray value between CBCT image and planning CT image, we normalized the pixel gray value of images prior to registration.In comparison to the CT image, the CBCT image has a poor quality. There are several factors affecting the CBCT image quality:truncated projection, beam hardening, the patient physiological movement, the compensation and gain correction of detector, the speckle, the misaligned CBCT geometric structures and so on. In order to improve image quality, and to reduce the irradiation dose of the normal tissue, Ravishankar N. Chityala proposed a scheme:reconstruct image in the region of interest and estimate image in the other area with a priori knowledge. The algorithm can remove truncated artifacts and reduce the irradiation of normal tissue. SEUNGRYONG CHO proposed the weighted density reconstruction algorithm of the region of interest. And this method can reduce the irradiation of the edge of the region of interest.In order to overcome the problem of blurred edges of the tumor target, Jenghwa Chang applied a respiratory gating technology to monitor the patient respiratory cycle. In radiation therapy, in the inhalation phase start radiotherapy and in exhaled phase stop radiotherapy or vice versa. The experimental result indicated that this method can improve image quality. We want to use the image of the same breathing phase to reconstruct CBCT images. And this can significantly improve the image quality. But the same respiratory phase projection slice data are too little and can not meet the traditional Shannon formula. So we propose a compression sensing theory applied to the CBCT image reconstruction algorithm, and give the experimental results. In addition, we descripted in detail the FDK reconstruction algorithm, the widely used parallel beam FBP reconstruction algorithm and the reconstruction algorithm in region of interest.In research of the CBCT image-guided radiotherapy system we studied the patient positioning error correction problem, the deformable registration problem between CBCT image and planning CT image, dose registration issue, clear image reconstruction problem. In order to solve these problems, we discussed the rigid registration transform, the deformable NURBS-based registration, the multi-resolution analysis, the optical flow field alignment algorithm, the compressed sensing theory etc. The main work in this dissertation includes the following aspects:(1) Discuss in detail the patient position error correction and gave the rigid registration algorithm between CBCT images and planning CT image and experimental result. We compare the performance of the ITK software and commercial the medical image processing software, MIM. We referred that the coincidence between tumor target centroid and the origin of the coordinate system have an important impact on patient positioning. We proposed the rotation of the gantry instead of the rotation around the Y axis, the rotation of the collimator instead of the rotation around the X axis and around the Z-axis. Thus we can achieve six freedom degrees for patient positioning, which will improve effectively the CBCT positioning accuracy;(2) Propose a new contour-based similarity metric algorithm. Before the introduction of the fast similarity metric algorithm, image segmentation algorithms are reviewed. We use the Snake model to segment the brain CT and MRI images. In experiment we compare the contour of fast similarity metric algorithm with the mutual information similarity metric algorithm. The experimental data show that the algorithm in this dissertaion presents a less number of iterations and the algorithm is feasible;(3) Introduce the multi-resolution analysis and the development of medical image registration. We study the deformable registration algorithms based on B-spline and NURBS-based, and give a detailed experimental result. The experimental result shows the registration error of the proposed NURBS based deformable algorithm is approximately3.5mm and approximately a third of common external margin in present. This algorithm improves highly the registration precision. This is significant for improving patient positioning accuracy and to reduce the irradiation of normal tissue around the tumor;(4) Introduce the dose-volume distribution and the definition of GTV, CTV, PTV, IV, OAR etc. Use the optical flow field model to register CBCT image and planning CT image. The experiment gives the optical flow field vector diagram and dose volume histogram of the parotid. This study is very significant for the precise CBCT image-guided radiotherapy system;(5) Introduce the imaging mechanism of CT and CBCT, the Radon transform and the projection theorem. We descript the FDK reconstruction algorithm and the parallel beam FBP reconstruction algorithm. In addition, we discuss the reconstruction algorithm of the region of interest. Because the projection slice data in the same respiratory phase is very little, Shannon formula can not be meted. We apply the compressed sensing theory to the CBCT image reconstruction algorithm, and give the experimental result.