CT/MR Image Registration Based On Contour Information

Radiation therapy is one of the main treatments for nasopharyngeal carcinoma.When developing a radiotherapy plan,the physicist delineates the target area based on CT images.However,CT images are not obvious for soft tissue,so that the target area is not accurate.Compared with CT images,MR images have more soft tissue information and can distinguish normal tissues around the target area.MR images are registered with CT images to compensate for the lack of CT images in soft tissue information,thereby increasing the preciseness of delineating the objective region.Most registrations require manual intervention to select the region of interest,which increases the workload of the physicist in a certain extent.At the same time,manually delineating the region of interest will lead to the randomness of the dimension of the target area,thus affecting the registration result and increasing the registration time.This thesis deals with the above problems in CT and MR image registration through different strategies.The major examination substance involve:(1)Segmenting on two-dimensional slices,and taking the region of interest(ROI)on the slice as the initial region of adjacent slices,continue to segment until the whole three-dimensional volume data is segmented.(2)Downsampling the CT and MR images to obtain multi-level resolution images.(3)Obtaining the corresponding contour features from the low-resolution CT and MR images,and performing surface matching on the contour features,and calculating the overlapping regions of the fixed image and the matched moving image according to the matching error,thereby automatically extracting the region of interest.(4)Using low-resolution images for surface matching,using the transform parameters obtained by matching to spatially transform the high-resolution images,and using the transformed transform parameters as initial parameters of rigid registration based on mutual information,and then using rigid-transformation parameters as initial parameters for non-rigid registration to obtain the final registration result.Ten sets of MR data are used to evaluate the segmentation algorithm.By comparing with experiments,the DICE and PM coefficients of the proposed segmentation method are mostly higher than those of Level Set algorithm.The segmentation results are in good agreement with the gold standard outlined by doctors.Fifty sets of CT and MR data are used to evaluate the registration algorithm.Through experimental comparison,the registration method proposed in this paper is superior to Maes’ mutual information method in registration effect,and reduces the registration time to a certain extent.