Research Of Radiation Therapy Image-Guided System

The rapid development of computer performance has brought tremendous help to many industries,especially in the field of medical image processing.In the current clinical application,medical imaging is the cornerstone of a large number of computer applications.It is used in the entire process of medical diagnosis,treatment planning and even surgical treatment.The medical image registration is the most basic step in medical image processing.It is a technique for spatially aligning two input images after a series of spatial transformations,and it's the focus of this thesis.The research points of this thesis come from two requirements in the horizontal project of the laboratory.The first research points is to monitor the posture changes of patients undergoing radiation treatment in real time to automatically stop the treatment when the patient has a large range of motion,and the second research points is to fuse the multi-modal medical images to help the doctor make a more accurate radiotherapy plan.The core of these two functions is image registration.At the same time,the image registration algorithm needs to be modified accordingly to adapt to their respective application scenarios.For demand one,its essence is a single-modal image registration problem that requires high real-time performance.It is roughly divided into two steps.First,roughly align the two-dimensional image in the application scene on the background,and then the magnitude of the patient's posture change is judged by measuring the similarity difference of the lesion area.Step one requires rigid registration,and step two requires the use of Similarity measure in image registration.In order to meet the real-time standard,this thesis has conducted in-depth research on several aspects of approximate normalized mutual information based on GPU-accelerated computing.At the same time,it uses the masking technology in image processing technology,and uses multi-resolution,multi-Level space conversion strategy,which ultimately improves the accuracy of the results and the calculation speed.For the multi-modality medical image fusion in the second requirement,it can also be divided into the multi-modality medical image registration problem and the registration-completed multi-modality medical image fusion problem.For this application scenario,this thesis focuses on the selection of transformation model,the selection of similarity measures and the image fusion algorithms.The combination implements a set of algorithms for rapid registration and fusion of multi-modal medical images,and implements a simple realization of the three-dimensional visualization of medical images.In order to ensure the real-time nature of the 2D image registration,for the timeconsuming iterative optimization process in image registration,an approximate normalized mutual information algorithm using GPU accelerated calculation is proposed,and it is applied to real-time In the application of monitoring the treatment process of patients;in addition,I reasonably select the components of the existing image registrationR algorithms and image fusion algorithms to achieve a set of multi-modal medical image rapid registration and fusion algorithms.In the end,tests show that the timeliness and functional requirements of the algorithm in this thesis meet expectations.