| Radiotherapy is a common means for treating lung cancer.In order to reduce the dose of radiotherapy,it's important to estimate the patient's respiratory motion via image registration of lung 4D CT(four-dimensional computed tomography).This article focuses on MRF-based deformable image registration techniques.A new image registration method for lung 4D CT based on high order power method(HOPM)is proposed.The main results are as follows:First,this paper introduces the mathematical model for MRF-based image registration,and then analyzes the effect of each potential function on displacement field respectively in terms of clique factorization.For 2D(two-dimensional)and 3D(three-dimensional)images,isotropic potential functions for topology preserving are presented.This paper also presents a general algorithm scheme of MRF-based image registration.Second,a tensor model is developed for image registration problems with high-order MRF on which some discrete optimization methods such as graph-cuts and belief propagation cannot effectively be applied.Instead,this paper shows that HOPM can be used to deal with the problem if the corresponding MRF is isotropic.To reduce the spatial complexity of the algorithm,this paper also presents a tensor sparse representation that can be used to construct tensor for(composite)homogeneous MRF.In addition,the potential function for weak topology preserving is improved for 2D images.The simulation results show that applying both smooth constraints and weak topology preserving constraints simultaneously on the displacement field can effectively preserve the topology of the field.Third,two kinds of multi-scale processing schemes are given for lung 4D CT image registration.The proposed method is evaluated on DIR-Lab public data sets.The impact of auxiliary functions is also carefully studied in this paper.Experimental results,with target registration error as a metric,demonstrate the feasibility and effectiveness of the method. |
PDF Full Text Request