Research On 2D/3D Medical Image Registration Technology And System Design In Image Navigation Surgery

The rapid development of medical imaging and computer science has not only brought convenience to modern surgery,but also brought new challenges.There are many kinds of medical images,with different imaging methods and image characteristics.For example,anatomical images extract anatomical morphological information of organs with high resolution,in which CT is more suitable for detecting dense tissues.Clinicians can quickly and accurately observe the location of focus points or other areas of interest through CT images,but they need to use CT machines for scanning imaging,which is not suitable for real-time imaging during surgery.However,two-dimensional medical images,such as X-ray,fast imaging methods just overcome this shortcoming.Therefore,in clinical practice,doctors often need to use a variety of medical images to register a variety of different modes of medical images.This technology is called multimodal medical image registration technology.Medical image registration technology is often used in image fusion and image navigation surgery.Image navigation surgery generally refers to minimally invasive surgery based on multi modality medical images and computer intervention.The key is medical image registration technology.Through medical image registration,images taken during surgery and images in surgery can be mapped to the same coordinate system or physical space.The most common registration is 2D/3D registration,that is,preoperative CT image registration and intraoperative X-ray image registration.In this paper,2D/3D medical image registration technology in image navigation surgery is studied.The main contributions are as follows:This thesis first investigates the field of 2D/3D medical image registration,analyzes its basic principles and framework,then improves the optimization algorithm,similarity measurement method and registration process,and finally designs a medical image registration system.For similarity measurement,this chapter focuses on the research of similarity measurement in medical image processing.Firstly,several common algorithms in gray based and feature-based classification are introduced and summarized according to their different characteristics.In view of the drawbacks of the two methods,and in order to better match the characteristics of medical images,this paper proposes a similarity measure algorithm based on image contour.Firstly,two medical images are preprocessed to filter most of the noise and interference information;After that,the main body contour information in the image is extracted.Finally,according to the distance between the corresponding contour points of the two images,the distance is weighted by Gaussian,and the ratio of the total weight value and the contour points is output as the size of the similarity measure.Finally,we designed a contrast experiment to verify the effect of the algorithm.The experiment shows that this paper can obtain more robust and accurate results in the medical image registration environment.For the optimization algorithm,the Equilibrium Optimizer is prone to uneven distribution and high randomness for population initialization.To solve this problem,this paper introduces the Logistic Tent chaotic map to replace the random initialization population,ensuring the dispersion of the initialization population and covering the entire search space as much as possible;The formula of iteration function in parameter update is modified to increase the randomness and step size of the first and middle stages,which is more conducive to finding the global optimal solution;Levy flight strategy is introduced to perturb stationary particles to prevent the algorithm from falling into local extremes.Then,benchmark function experiment and simulation registration experiment are designed to verify the improved algorithm.The results show that the improved Equilibrium Optimizer can obtain more accurate results than other optimization algorithms.The registration system equipped with the improved Equilibrium Optimizer can obtain better registration accuracy.Later,in the registration system,the speed is reduced due to the frequent data transmission at the Host and Device during the transmission of DRR images.Single data is batched into data groups,and they are transmitted in parallel at the Host and Device ends,thus reducing the number of data transmissions and thus reducing the registration time.Finally,comparative experiments are designed to verify the acceleration effect of the improved CUDA parallel execution process,and the results show that the registration speed can be significantly improved.Finally,based on prior knowledge of medical image registration and the algorithm framework described in this thesis,a medical image registration system was built using C++,Qt,CUDA,and ITK.The system consists of three layers: the user interface layer,the algorithm layer,and the data layer,and it is capable of performing image registration tasks.Simulated registration experiments were designed to validate the system’s stability,accuracy,and time consumption.Experimental results have demonstrated that the system described in this thesis is able to meet the basic requirements of real-time intraoperative registration.