Research On Algorithm Of Lung Image Registration Based On Mean Square Difference

The lungs are the important respiratory organs of the human body.In the diagnosis of lung diseases,doctors usually need to use lung imaging equipment to assist in the diagnosis of lesions.Sometimes it is necessary to compare and analyze the lung images obtained from different periods.First,they need to be aligned in the space position,that is,image registration.Lungs inevitably produce complex deformations due to respiratory motion.Therefore,registration of lung CT images requires non-rigid registration algorithm.Lung image registration is a difficult topic in the field of non-rigid registration.It is also one of the key steps in lung image analysis,which is of great significance.In this paper,the single mode CT image of lung is used as the data source,based on the similarity measure of mean square difference and Gauss Newton L-BFGS(Limited-memory BFGS)optimization algorithm,the problems of large computation,local extremum and poor registration precision are studied.An improvement is put forward on the key steps of similarity measure,space transformation,optimization algorithm and so on,A non-rigid registration method for lung CT image is proposed.The experimental results show that the method has achieved good results.The main contents and contributions of this paper are as follows:1.In this paper,a new single mode non-rigid registration algorithm based on improved mean squared difference is proposed on the basis of the study of the mean square difference measure and the penalty term.This method improves the registration accuracy.Compared with the traditional mean square difference measure,the registration accuracy of the improved mean square difference has been greatly improved in the lung CT image.2.In this paper,a single mode non-rigid registration algorithm based on mean squared difference with regularity of FFD is designed for the severe deformation of the lung.In view of the serious lung deformation,the FFD model is used to simulate the lung deformation,and a hierarchical spatial transformation model is introduced.The registration process is divided into coarse registration and fine registration.Coarse registration uses affine transformation to align the image globally,and then performs fine registration.That is,the FFD model is used to correct the local shape of the image.In the fine registration,to balance the accuracy and computational complexity of the FFD model deformation,a multi-resolution strategy was introduced to change the number of control points layer by layer.And use the mean squared difference with regularity as the standard to measure whether the image is aligned.The regularity is to control the smoothness of deformation.Compared with the improved mean square difference measure and the traditional mean square difference measure,the registration accuracy is greatly improved.3.For the L-BFGS optimization algorithm is easy to fall into the local extreme characteristics,a hybrid optimization algorithm based on the combination of L-BFGS and CMA-ES is proposed.During the registration process,the CMA-ES algorithm is used for global search first,and then L-BFGS algorithm is used for local search.Experimental results show that the hybrid optimization algorithm proposed in this paper has achieved good results.