A multimodality image registration system in radiological science : with special consideration in clinical applications

Abstract

This thesis appraises the development and assesses the methods for registration of medical images from multi-imaging modalities. In particular, problems of deformable anatomical structures are addressed. Different methods of image registration of deformable structures are evaluated. A warping-deformable model is proposed to tackle the problem of registration of CT (computed tomography) and MR (magnetic resonance) images of the abdomen with the liver as a target organ. The model can be extended to different imaging modalities and deformable structures alike. A two-step strategy is suggested where the global transformation is firstly used to determine the global mapping and in the second step, local 7ilarping is implemented for the local target organs by radial basis function. Eight sets of CT and MR images of the abdomen were used to evaluate the performance of the model. The accomplishment of matching is measured using statistical method (mean and standard deviations) and cross correlation coefficient of the matching images. The paired t-test of a small sample indicated that the transformed image using the warping-deformable model is significantly better in performance than those images without transformation. This study suggested the anatomical landmarks specific for image registration of the abdomen and, specifically, the liver. The study also demonstrated that adequacy of the warping-deformable model for registration of deformable human organs and the technique developed here can be extended to other imaging modalities, especially ultrasound.