Non-rigid Multi-modal Medical Image Registration Based On Modal Reduction And Conversion

A variety of medical imaging modalities have emerged with the development of medical technology.Clinicians can use medical images for disease diagnosis because different imaging modalities can provide different physiological information.Medical image registration is an important part of image processing and computer vision.It is also a key prerequisite for accurate image fusion.Multi-modal medical image registration is beneficial to complement the information between different modal images,and the information complementary images provide a variety of information of diseased tissues or organs,providing a powerful theoretical basis for doctors to make accurate diagnosis.This paper mainly focuses on the exploration of non-rigid multi-modal brain images,hoping to effectively promote the theoretical research and application of medical image registration.The main work and innovations of this paper are summarized as follows:Concern the problem that noise and intensity distortion exist in brain images,the method based on structural information can not accurately extract image intensity information,edge and texture features at the same time.In addition,the computational complexity of continuous optimization is relatively high.According to the structural information of the image,non-rigid multi-modal brain image registration method based on improved Zernike moments based local descriptor(IZMLD)and graph cuts(GC)discrete optimization was proposed.Firstly,the image registration problem was regarded as the discrete label problem of Markov random field(MRF),and the energy function was constructed.The two energy terms were composed of the intensity similarity and smoothness of the displacement vector field.Secondly,a smoothness constraint based on the first derivative was used to penalize sharp changes in the adjacent displacement labels across pixels.The similarity metric based on the IZMLD was used as a data item to represent intensity similarity.Then,the Zernike moments of the image patches was used to calculate the self-similarity of the reference and the floating image in the local neighborhood and construct an effective local descriptor.The sum of absolute difference(SAD)between the descriptors was taken as the similarity metric.Finally,the whole energy function was discretized and the minimum was optimized by using the extended optimization algorithm of the GC.The experimental results show that the proposed method achieves efficient and accurate registration when image has noise and intensity distortion.Concern the problem that the anatomical information may be lost during modal conversion based on structural information method,and the single direction image synthesis for image registration causes the deviation due to ignoring the anatomical details in the other modality.In this paper,we propose a bi-directional image synthesis based approach for MR-to-CT non-rigid medical image registration.First,a simple pre-registration was performed on the reference CT image and the floating MR image.Then,bi-directional image synthesis was performed using two improved random forest with auto-context optimization model.Specifically,the multi-target regression forest algorithm was used to synthesize CT,and the multi-level multi-target regression forest with stronger learning ability was used to synthesize MR(S-MR).Finally,a dual-path fusion framework was developed to iteratively and effectively combine two registration pathways to estimate the deformation path between CT and MR images.Two deformation pathways: 1)one from the S-CT to the actual CT and 2)another from actual MR to the S-MR.Dual-path fusion framework by using complementary information from both modalities.The experimental results show that the use of bi-directional image synthesis for image registration improves the accuracy of non-rigid multi-modal medical image registration.Aiming at the two improved methods based on modal conversion proposed in this paper,a non-rigid multi-modal brain image registration system was developed based on MATLAB programming platform.The system was mainly divided into two modules: structural representation method and bi-directional image synthesis method.Compared with the traditional method,the method proposed in this paper has better performance.At the same time,the stability and practicability of the system are demonstrated,which can be used in practical clinical diagnosis.