Research On Brain Mr Image Registration Based On Gyral Hinge And 3D Res-Unet

Magnetic resonance imaging(MRI),as a common technique of assistant diagnosis,has been widely used in the field of brain imaging.The formation of MRI will be different because of different radio frequency pulse sequence,so if there are differences in equipment parameter setting and radio frequency pulse,the formed magnetic resonance image will be different.Fusion of different brain MRI has important reference significance in clinical practice.Before image fusion,image registration operation must be performed and plays an important role.At present,with the continuous enhancement of hardware capability and the outstanding performance of convolutional neural network,the medical image registration algorithm based on deep learning has become the mainstream of research.However,it is difficult to label the transformable deformation field manually,resulting in the lack of real label data,so the supervised deep learning method is facing difficulties in practical application.Based on this,this thesis employs the unsupervised deep learning method to carry out research.At the same time,considering that the cerebral cortex is highly folded,small change and complex structure,we introduce an image feature mapping – gyral hinge to assist the registration process and improve the registration accuracy.The main work of this paper is as follows.1.Gyral hinge is introduced to assist registration.Considering the structural characteristics of cerebral cortex,gyral hinge can effectively express the threedimensional spatial information of the image with a small amount of data,and has a certain consistency in different brains.We extract the brain MRI image mask-gyral hinge as an auxiliary feature for registration,and improve the registration performance without increasing the amount of data as much as possible.2.The registration procedure is performed in a novel unsupervised learning manner.In this paper,an unsupervised learning algorithm for brain MRI image registration based on gyral hinge is proposed.The core of the algorithm is to construct a three-dimensional U-net based on long and short connection residual structure.By using spatial transformer transformation module,the learning deformation field can be directly applied to the image.At the same time,the loss function is constructed by introducing Jacobian loss matrix.So that the deformation field is more accurate and the overlap is less.LPBA40,IBSR18,MGH10 and CUMC12.The experimental results verify the effectiveness of the method.And the effectiveness of the proposed algorithm and related super parameters are analyzed in the final.3.Experiments are conducted on four widely used medical registration datasets,LPBA40,IBSR18,MGH10 and CUMC12.The experimental results verify the effectiveness of the method.And the effectiveness of the proposed algorithm and related super parameters are analyzed in the final.