A Medical Image Elastic Registration Approach Based On Radial Basis Functions And Partition Of Unity

Registration of medical image is an important subject in medical image processing. The past researches mainly use linear-transformations to make the registration, these methods don't work well on the images which have nonlinear-deform in some places, so the nonlinear-transformations become the key issue. Previous work has concentrated on selecting the corresponding landmarks manually and using thin-plate spline to gain the global elastic transformation.Because the involved matrix is dense, we couldn't select enough landmarks which assure the accuracy, or else the time cost is unbearable.After comparing the properties and the interpolation effects of different radial basis functions ,combining the partition of unity method,this paper proposes a fast evaluation of radial basis functions method,and discussed its approximation property.Based on the researched above,this paper introduces an algorithm which could uses large-scale landmarks to gain a higher registration accuracy:first of all,we develop the traditional registration scheme to let it,adapt to all radial basis functions application;and then by using thinning algorithm, we get the landmarks which are large-scale and important ;then give a covering of the original image, choose an approximation space for every cell;at last put the local approximants together by using the partition of unity method.Experiments show that it is an effective and accurate approach,which also has adaptive application potential.