| In recent years, medical image registration is used in many fields, such as surgicalnavigation, motion estimation of organ or tissue, radiation therapy. Nowadays, it hasbecome a hotspot in medical image process field for researchers. Medical imageregistration is such an important technology in these fields. By this method, thedifferences between the location of patient’s tumour and position of radiation beam aredetected and this process is in real-time. These differences involve setup errors, organmotion and so on. The purposes of registration are adjusting the position of treatmentcouch or the radiation in real-time, decreasing the spatial difference and protecting thenormal tissue. Using registration between two-dimension (2D) images and that ofthree-dimension (3D) in medical field can compensate the position differences, and thistechnique improves the precision of radiation therapy in a large part. Therefore, how toimprove the accuracy of registration on the premise of fast speed becomes meaningfulduring radiotherapy process.In this dissertation,2D-3D medical image registration is mainly applied toradiotherapy system. It adopts X-Ray images and planning CT for registration. In orderto calculate the registration parameters accurately, an improved registration algorithm isproposed. In this method, the six transformation parameters are decomposed intoregistration parameters of two orthogonal plains. Rigid transformation in patientcoordinate is divided into two orthogonal directions. The process in each directionis segmented into registration in plains and that is not in the two plains. The rigidtransformation parameters are calculated with the registration results in the twodirections by2D-3D geometric transformation. Mutual information and modifiedPowell search method are proposed to use in some parts of the algorithm to improve theperformance. Given these parameters, the radiotherapy system adjusts the position toassure accurate radiotherapy.Before the registration process, Digitally Reconstructed Radiograph (DRR) isgenerated firstly and the results show that DRR in the two orthogonal directions from3D CT data are quite good. These data are prepared for the subsequent process. In thisdissertation, image registration will be separated into four steps. In the first step,pyramid strategy is used. Fast mean square gray difference is also used as similarityfunction to obtain the two sets of initial image registration parameters in two orthogonalplanes. The second step is to achieve the initial parameters of spatial transformation.One-dimensional Brent search method is proposed to use during the process ofregistration between non in-plane DRR and X-Ray images. The third step is using amore precise similarity function (mutual information) and search algorithm. Partialvolume, mutual information and improved Powell algorithm are proposed to optimize the initial results. In the fourth step, according to the previous registration results, wecan calculate the transformation parameters between3D CT data (patient position) andX-Ray images (radiation energy beam) by geometric relationship between the twosystems to achieve the purpose of image guided radiotherapy.The mutual information method with partial volume interpolation that is used inthis dissertation has more obvious peak and better anti-noise performance, comparedwith pattern intensity similary function from a literature. In the aspect of searchingoptimal values, our design uses the modified Powell method. It has faster search speedand higher search accuracy. Our experiment adopts a set of CT images to obtain twosets of DRR, and two virtual X-Ray images are generated by making some spatialtransformation of two DRR images. The simulation platform of the experiment isMatlab2010. The experiment results show that the improved registration algorithm hashigh accuracy. The translation errors are in one millimeter or less. The overallregistration accuracy reaches sub-pixel level in the premise of the speed withinallowable range. |
PDF Full Text Request