Research On Non-rigid Lung Medical Image Registration Algorithm

Lung is a typical locomotive organ. Lung imags are greatly different in the effect of acquisition environment, accuracy of equipments and respiratory movement in the lung image acquisition process. It is necessary to correct the whole and local deformation in lung images to avoid misdiagnosing and missed diagnosis in the diagnosis of lung disease, treatment and disease tracking. Therefore, accurate correcting lung deformation and realizing pulmonary nonrigid registration have become a key technology to be solved at present. The theory of medical image registration is summarized, and the development history, significance and difficulties of research are introduced in brief. Afterwards, we studied the customized registration algorithm facing different clinical needs. This paper focuses on non-rigid lung image registration algorithm, the registration between lung PET-CT images and the registration on lung inspiration and expiration images.We propose the MIGTP demons image registration method aming at lung PET-CT image registration. Active Demons model can perfectly correct the nonrigid deformation, but there are two problems in the model. Firstly, it only can correct the nonrigid deformation of monomode images, and can’t be used in multimode registration. Secondly, the false demons force exists in many cases. So, the topology structure of registered moving image is easy to tear out, deform and fold during registration. This paper proposes a mutual information gradient and topology preserving demons model, which can not only register the multimode images but also align the nonrigid deformation. Mutual information criterion allows for robust and automatic registration without any prior precessing steps. The mutual information gradient based on Kullback-Leiber distance is added to the Active demons model to make the moving image change in the mutual information increasing direction, which make the model register multimode images. Moreover, the intensity distribution of moving image and static image is analyzed. Then, the topology preserving method is added to the Active demons force to keep the topology structure of moving image. To test the parameter effect in the model, some simulations are made. The experiment result shows the best values of these parameters.The image segmentation method and b-splines deformation model are combined to correct the deformation of inspiration and expiration lung images. B-splines deformation model can accurately algin the nonrigid deformation of images, but the computation complexity is high. The registration of segmented lung image can avoid calculating some unuseful data. The simulation results show that the registration accuracy of the proposed algorithm is high.