Research On Medical Image Registration Algorithm Between CBCT And PCT Based On Super-resolution

Smart healthcare and automated healthcare have become the theme of the moment with the development of technology.The advent of various types of medical imaging has made Image-guided Radiation Therapy（IGRT）an important tool in the treatment of tumors.IGRT observes the tumor condition through the registration of medical images before and during radiotherapy of tumor patients,accurately locates the tumor location and protects normal tissues.It is a key technology for tumor treatment.Cone beam computed tomography（CBCT）registred with Planned CT（PCT）which is the most common modality used to treat tumor in IGRT.However,CBCT carries artefacts,poor image quality and low registration accuracy when registred with PCT,which can affect medical diagnosis and treatment.For this reason,this paper addresses the issues of CBCT image quality and the accuracy of the registration algorithm of CBCT and PCT to improve the tumor cure rate.The research content is as follows.The Super-Resolution（SR）image enhancement algorithm is proposed for the quality problem of enhancing the soft tissue part of CBCT with low contrast and easily with noise.SR algorithm image enhancement is first performed in processing the unified dataset CBCT.The experiments were designed to compare the peak signal-tonoise ratio（PSNR）index data with the bicubic interpolation,the average PSNR after bicubic interpolation was 37.24 db,and the average PSNR of SR algorithm was42.12 db.Then the effects of registration accuracy with and without SR algorithm enhancement on CBCT and PCT are compared on three rigid registration experiments,and the results demonstrate the superiority of SR algorithm enhancement.A new measure function called S-PCC,which is a combination of structural similarity and Pearson correlation coefficient,is proposed for the problem of similarity measure function of medical image deformed registration model.The image deformable registration experiment uses mean square error,mutual information,structural similarity and Pearson correlation coefficient as registration measures for CBCT and PCT,respectively,and then compares the new measure function S-PCC registration effect.The SR algorithm enhancement in Chapter 3 is then combined with the new measurement function in the deformed registration model to compare the effect of the presence or absence of SR on the registration accuracy.Aiming at the problem of incomplete information about the image organization structure of the medical image registration data,this paper converts the medical image into a three-dimensional image format to make up for the missing structure of the twodimensional image in the image registration and the associated information of the upper and lower slice image positions.The deformed registration model makes full use of image information.In this paper,the registration experiment of comparing twodimensional and three-dimensional images as data sets shows that the registration accuracy of three-dimensional images is higher.