Research On Medical Image Registration Based On CNN

In clinical medicine,it is often necessary to compare and analyze two or more images to reach a conclusion.However,due to the medical image data source is various,the time is inconsistent,and other factors,which make two images can not do one-to-one correspondence.Medical image ratio is easily realized by image registration technology,which plays an important role in clinical medical diagnosis,and the depth learning method improves the accuracy and speed of medical image registration.In this paper,medical image registration based on CNN is studied.The main results are as follows:This paper proposes a registration model based on Voxel Morph model.The full-scale jumping connection of the coding process is added to the registration network in the Voxel Morph model,which makes the decoding process retain more features and information to more accurately predict the deformation field.The registration model is LDIRnet registration model,which connects two improved registration networks in series.Each improved registration network uses the unsupervised learning registration method to predict a small deformation field,and the last two small deformation fields are superimposed to obtain the final deformation field.The model is evaluated by the oasis datasets(T1-weighted MRI images of the brain),Wavelet decomposition and reconstruction are also used to enhance the image before model training.Compared with the affine transformation method,the Voxel Morph-1 model and the Voxel Morph-2 model,the mean DSC coefficients of the proposed method is increased by 28.6%,1.2% and 0.2%.The experimental results show that the method in this paper can improve the accuracy of image registration effectively.The registration model is improved.Not only the full-scale jump connection of the coding process is added to the registration network,but also the fixed image and the moving image are added in the coding process.This model can not only make the coding process capture more image features,but also make the decoding process retain more image information and predict the deformation field more accurately.The model is evaluated by the oasis dataset(T1-weighted MRI images of the brain).Compared with the affine transformation method,the Voxel Morph-1 model and the Voxel Morph-2 model,the mean DSC coefficients of the improved model is increased by 28.9%,1.5% and 0.5%.The experimental results show that the improved model has better registration effect.