Research Of Medical Image Registration Method Based On SNBC And APSO

With the development of medical imaging, medical image registration has become one of the topics. Medical image registration has significant clinical value. It is widely used in the areas of clinical diagnoses, treatment, surgery navigation and evaluation of curative effect. In recent years, rigid registration technology matures with each passing day and perfect, non-rigid registration of medical image is a more important research topic.In the method of feature-based non-rigid medical registration, the formation of matching points is one of the essential steps. In order to make image matching approach is more efficient, the paper presents Semi Naive Bayesian Classification (SNBC) mothod to realize feature matching. The algorithm first detects interest points in reference image by rapid Hessian detector, constructing training set, then classifier is obtained by training phase, and then generates the final match points using SNBC method. The algorithm computational cost is low in the classification phase.In the paper, multilevel B-spline is used as transformation model. It makes control lattice change form coarse to dense. The shape feature is obtained by coarse lattice, while finer lattice is helpful to reduce the approximation errors to register two images accurately.Optimization algorithm in image registration is mainly used to search the optimal parameters of spatial transformation, which performance has a direct impact on the registration result. Particle Swarm Optimization (PSO) algorithm is widely used in practical optimization problems because of its simplicity and parallel computing. However, PSO is easy to fall into local extremum with low convergence speed. Because of this, Adaptive Particle Swarm Optimization (APSO) method is used in the paper, and partial Hausdorff distance is selected as the similarity measure. Experimental results indicate that APSO can converge to the global optimum at faster rate in the whole search space.A software platform of non-rigid image registration has been designed based on OpenCV and VS2005. A large number of registration experiments have been carried out in the platform. Compared the algorism proposed in the paper with the classical Scale Invariant Feature Transformation (SIFT) operator and PSO algorithm, the experimental results show that the algorithm based on SNBC matching and APSO is more superiority.