Study And Application Of Non-Rigid Image Registration Algorithms In Focused Ultrasound Therapy System

Focused ultrasound therapy technique has been used on clinic as a new method of tumor therapy about ten years. In recent years, it rapidly developed, and a great deal of experience has been accumulated on the related basic research,the device manufacture and the clinical practice. The therapeutic efficacy of focused ultrasound operation is decided by accurate localization of the tumor target. At present, most of these equipments use B-mode ultrasound to guide the operation. Their therapeutic efficacy depend on the ability and experience of clinical doctor, which hampers the ultrasound thermal therapy developed into the accurate treatment in a certain degree. Currently, the most important thing is pursuing the accurate planning,the accurate localization and the accurate therapy for increasing the efficacy of thermal therapy. The magnetic resonance (MR) image is used to guide the operation in ideal condition, but the price of MR imaging device is higher than ultrasound device, and installing MR in focused ultrasound system requires some essential techniques. It is practical and imperative for the accurate therapy to search the straight,effective localization and monitor method based on the intra-operative B-mode ultrasound.Elicited by the advancing localization and monitor technique in Image-guided Radiotherapy (IGRT), we make use of pre-operative CT/MR images and develop image registration methods are suitable for focused ultrasound therapy system, to achieve the accurate localization and monitor in focused ultrasound operation on the basis of intra-operative B-mode ultrasound imaging. This dissertation contains the following research results:(1) Recent developments in focused ultrasound system and its localization,monitor technique are reviewed. After the comparison and analysis of localization and monitor technique between focused ultrasound therapy and radiotherapy system, the non-rigid registration methods appropriate for focused ultrasound therapy system are presented. Image-guided localization and monitor in focused ultrasound therapy system is implemented by using the pre-operative CT/MR image of patient and modified registration technique based on existing B-mode ultrasound imaging. The key of function of image-guided localization is 3D registration and fusion between pre-operative CT/MR image and intra-operative ultrasound image. The emphasis of function of image-guided monitor is on 2D non-rigid registration of intra-operative ultrasound images.(2) Free-Form Deformation (FFD) model is able to describe the local deformation of 3D deformable objects very well in medicine. Via study of FFD model based on B-splines and mutual information, a new interpolation algorithm (called HPV) that uses an approximate function of Hanning windowed sinc as kernel function of partial volume (PV) interpolation is presented, in order to reduce the local extreme as a result of traditional interpolation methods. In addition, a new method is given to estimate the gradient of cost function with respect to the FFD model parameters based on HPV interpolation. The new non-rigid registration algorithm is composed of the B-splines FFD model,the integrated metric including mutual information and smoothness constraint function,HPV interpolation and gradient descent optimization method, which is applied to the 3D registration and fusion between pre-operative CT/MR image and intra-operative ultrasound image. By both synthetic and real data experiments on CT and MR images, the new algorithm shows the ability to restrain the local extreme.(3) Because physical model can keep the topological properties of deformation and thus constrain the enormous solution space, it is used to model the breath motion for intra-operative ultrasound images. It is verified that linear-elastic model is derived assuming small and linear transformations and has small computational load, but viscous-fluid model accounts for the large distance deformations,is homeomorphic and has complex computation after discussing physical model-based non-rigid registration method. In order to improve the accuracy of registration results, the extraction of feature points representing shape and Scale Invariant Feature Transform (SIFT) method are also analysed. Especially for some small structure features, a new non-rigid registration algorithm incorporating shape information and SIFT based on fluid model is developed. For registration of the monomodal images of ultrasound, the new algorithm combines intensity difference metric,global feature points representing shape information and local keypoint extracted by SIFT in a Bayesian framework. From the comparison of the experiment results, the new algorithm is shown to improve the accuracy of registration.(4) Finally, the new non-rigid registration algorithm based on FFD model with HPV interpolation and the new fluid non-rigid registration algorithm is respectively applied to focused ultrasound therapy system, to implement the function of image-guided localization and monitor. The software system of localization and monitor is developed, which contains the three-view interface and 3D surface reconstruction of target by Microsoft Visual C++ platform,VTK and OpenGL software toolkit. By experiments on clinical data, the clinical validation of this function is verified.