Study On PET/CT Image Registration In The Chest And The System Design

With the development of the biomedical engineering technology and computertechnology, the medical imaging offers a variety of medical images modals forclinical diagnosis. PET/CT scanner is the most advanced equipment for cancerdiagnosis and radiation therapy especially the early diagnosis of the tumor. SinglePET image and single CT image can be acquired at the same time with the image afterfusion. As a result, there is misregistration in global and deformable errors in localcaused by respiratory movement or organ peristalsis, though PET image has beencorrected by CT-based attenuation. In order to guarantee the accuracy of diagnosis,the precise registration of CT image and PET image from PET/CT scanner isextremely important.The frame of image registration and some theoretical knowledge are presented inthis thesis, including similarity measures, geometric transformation, optimization,interpolation, multi-resolution registration and evaluation method.Redundant information in image registration process will increase the complexityof the calculation and affect the registration accuracy. In order to facilitate registration,PET and CT image are preprocessed respectively based on their characteristics. Bspline interpolation method is used to improve the resolution of PET image andmorphological transformation is used to remove the scanning bed information of CTimage. The registration methods proposed in this thesis include rigid registration andnon-rigid registration. The premise and velocity of different optimization algorithmfor rigid registration is researched using mutual information as the similarity measure,affine transformation as geometric transformation, including particle swarmoptimization, simplex optimization, genetic optimization and gradient descentoptimization. The results prove that different optimization has their own advantageand disadvantage in registration accuracy and velocity, but affine transformation can’thandle the deformation. Non-rigid registration method based on multi-transformationand multi-optimization is proposed in this thesis. Rigid transformation is used first toensure the global registration with simplex optimization and then BSplinetransformation is used for handling local deformation with LBFGSB optimization.The method proposed can handle the deformation and correct the registration errorcaused by respiratory motion, and is better than these methods using single transformation and single optimization.Medical image processing and image registration is very important in the field ofmedical image analysis. It is helpful to improve the efficiency of scientific researchworkers and medical staff to design a set of practical medical image registrationsystem for registration visualization. A set of medical image processing andregistration system based on the above image processing algorithm and registrationalgorithm is designed combined ITK, VTK and Qt. This system can achieve reading,storage and displaying of two-dimensional and three-dimensional and sequencemedical images. It contains a lot of functions about image processing, includingfiltering, edge detection, image morphology transformation, resampling, formatconversion, etc., and can undertake multimodal image registration including rigid andelastic registration with displaying the parameters after registration. The user canchoose different methods and parameters according to different situation. The systemhas a simple, friendly user interface, and is suitable for medical imaging and scientificresearch workers.