Research On Medical Image Registration Based On Multi-Scale Information

The significance of the people’s health problems are getting particularly important with the rapid development of society and the accelerating pace of work.When people seek medical treatment,the doctors often need to make correct diagnosis through the medical images of patients.Therefore,the research of technologies of medical image processing has been become a research hotspot in the medical field.However,the medical images’ comparison has been played an important role in clinical diagnosis and treatment,which includes the comparison between the medical images of different patients,or the comparison between medical images of the same patient at different times or different modalities,and so on.The paper focuses on medical images registration based on the multi-scale information of medical images,and the main tasks are as follows.For the rigid registration of medical images,a hierarchical stepwise approximation registration method for medical images is proposed.This method mainly consists of two steps.Firstly,the edge contour features of the two images are extracted.The principal axes method is used to complete the rough registration based on contour features.Secondly,the simplified multi-scale decomposition algorithm is used to extract information on different scales of the images,which achieves the diversity of extracted features.The feature information of the image is fully used for layer-by-layer approximation and registration.The registration process will start from the last layer,which upward registration layer by layer.Then the final registration result is obtained.The experiments show that the ideal registration result can be achieved by this method for complete or incomplete medical images.For the non-rigid registration of medical images,a layer-by-layer approximation registration method based on Thin-Plate Spline is proposed.The simplified multi-scale decomposition algorithm is used to obtain the sub-band information of the image to be registered at different scales firstly.Then these sub-band information’s non-rigid registration with layer-by-layer approximation are accomplished by Thin-Plate Spline function,i.e.Thin-Plate Spline function is used to accomplish the registration of sub-band information in the last layer of the target image and the reference image.And the bicubic spline interpolation function is used to reconstruct the target image after the registration,which gets the subband image of the previous layer.The reconstructed target image is used as registration object based on the Thin-Plate Spline function.The registration mode is executed until the end of the original image.The experimental results show that better results can be achieved by this method for the non-rigid registration of medical images.