Non-rigid Multimodal Medical Image Registration Based On Image Synthesis

Medical image registration technology is a very important link in the field of medical image processing,and it is also the basis for accurate image fusion.Multi-modal medical image registration complements the information between different modal images,and can simultaneously provide doctors with functional information and anatomical information of the lesion site,and provide a powerful basis for assisting doctors to accurately diagnose diseases.This paper focuses on the systematic research on image registration methods based on image synthesis.The work in this paper mainly includes the following aspects:Aiming at the problems of poor robustness of the synthetic model and insufficient representation of the feature information of the synthetic image in the existing image synthesis-based registration algorithms,a multimodal brain based on the generation of adversarial networks based on the dense relative average conditions of residuals was proposed.Image registration method.First,combining the relative average discriminator in the relative average generation adversarial network can enhance the stability of the model and the advantages of adding condition variables to the conditional generation adversarial network can improve the quality of the generated data.The ability to fully extract deep network features using the dense block of residuals to construct the RD-RaCGAN synthetic model.Then,the reference CT and floating MR images to be registered are bidirectionally synthesized by the trained RD-RaCGAN synthesis model to the corresponding reference MR and floating CT images.The region adaptive registration algorithm is used to select the key points of bone information from the reference CT and floating CT images,and the key points of soft tissue information from the floating MR and reference MR images.The extracted key points are used to guide the deformation field estimation.A deformation field is estimated from the floating CT image to the reference CT image.Similarly,a deformation field is estimated from the floating MR image to the reference MR image.In addition,the idea of layered symmetry is used to further optimize the two deformation fields.When the difference between the two deformation fields reaches a minimum,the two deformation fields are merged to obtain the final deformation field,and the deformation field is applied to the floating image to complete the matching.quasi.The experimental results show that the multi-modal brain image registration method proposed in this paper has strong robustness and can perform image registration tasks stably and accurately.In view of the limited scalability and robustness of image synthesis-based registration methods when processing more than two image domains,this paper proposes a medical image registration method based on edge-sensing multi-domain adversarial networks.First,combining the advantages of StarGAN v2 with style coding in a single generator to achieve multi-domain image synthesis and Sobel edge detection operators,the ability to fully extract deep network features using residual dense blocks to build a synthesis model.On the basis of improving the efficiency of multi-domain image synthesis,the integrity of the structural information of the synthesized image is guaranteed.Then,the reference CT and floating MR images to be registered are synthesized with the corresponding reference MR and floating CT images through the trained synthetic model.Finally,the region adaptive registration algorithm is used to select the key points of bone information from the reference CT and floating CT images,and the key points of soft tissue information from the floating MR and reference MR images.The idea of layer symmetry is further optimized.Finally,the deformation field is applied to the floating image to complete the registration.The experimental results show that this algorithm improves the practical application value of medical image registration algorithm in clinical medicine to a certain extent.Aiming at the two image registration methods based on image synthesis proposed in this paper,a non-rigid multimodal brain image registration system was developed in combination with the PyCharm platform.The system is mainly divided into two modules: registration method without image synthesis and registration method based on image synthesis.The registration effect of the traditional method is compared with the method proposed in this paper to verify the advancedness and robustness of the algorithm in this paper.At the same time,it demonstrates the stability and practicality of the system,which can be used to assist medical diagnosis.