Research On Elastic Registration Algorithm Of Medical Image Based On Spline Transformation

With the development of medical imaging technology, medical image has gradually become an important basis of the clinical diagnosis and treatment. The elastic registration of medical image with larger deformation is an important research topic. Wherein, the elastic registration algorithm based on spline transformation describes elastic deformation of the floating image by control points and spline basis functions. It has many characteristics. Thus, in this thesis, elastic registration algorithms are divided into three categories. They are elastic registration algorithms based on feature, gray and the combination of feature and gray. And on this basis, thin plate spline and B-spline are selected to perform spatial transformation. It achieves the registration of medical image with larger deformation quickly and accurately.In the feature-based elastic registration algorithm, SIFT algorithm cannot extract feature points in the larger deformation region. In order to overcome this shortcoming, an elastic registration algorithm based on progressively fine feature extraction and spline transformation has been proposed. First, SIFT algorithm is used to complete preliminary feature extraction and matching for the reference image and the floating image. It obtains feature points to reflect the global deformation. Next, the local region with larger deformation is obtained by calculating mean square difference of the pixel neighborhood. Then mutual information is utilized to complete feature extraction and matching in these regions. It obtains feature points to reflect the local deformation. Experimental results demonstrate that the algorithm reduces the calculating time of feature extraction and matching, and improves registration accuracy.In the gray-based elastic registration algorithm, each layer of the control grid is implemented global refinement by hierarchical B-spline transformation. It causes unnecessary calculations and affects the registration region. To overcome this shortcoming, an elastic registration algorithm based on hierarchical B-spline with local region refinement has been proposed. Firstly, the larger control grid is employed to complete global registration of the floating image quickly. Then the smaller control grid is used for the larger deformation region to achieve local registration accurately. Experimental results demonstrate that the algorithm reduces the calculating time of spatial transformation, and ensures registration accuracy.In order to achieve the registration of medical image with larger deformation quickly and accurately, an elastic registration algorithm with hybrid spline transformation based on feature and gray has been proposed. The algorithm combines the feature-based and the gray-based elastic registration algorithm. And it uses the advantages of thin plate spline transformation and B-spline transformation. First, the preliminary registration image is obtained by thin-plate spline transformation. And the input of this process is the feature point obtained by SIFT algorithm. Then mutual information is used as similarity metric. The preliminary registration image is transformed to obtain the final registration image by hierarchical B-spline with local region refinement. Experimental results demonstrate that the algorithm improves the speed and accuracy of registration.