Improved B-spline Elastic Registration for Medical Image Based on Feature Extraction

Abstract

In order to improve the accuracy and efficiency of medical image registration, an elastic registration method based on feature extraction and B-spline has been proposed in this paper. The proposed method consists of two steps. First, feature points are obtained by using Harris corner detection algorithm and SIFT algorithm. They are used for the input of Thin Plate Spline method to reduce the scale of deformation by preliminary transformation. It is beneficial to improve the speed of B-spline registration, which is used in the second step of the proposed method. Then the deformation grid is treated as the initial grid of B-spline free-form deformation (FFD), and accurate registration is achieved by B-spline transformation of mutual information (MI) driving. Experimental results demonstrate that the running time and registration results of the proposed method are superior to those of B-spline registration due to the combination of the efficiency of preliminary transformation and the accuracy of B-spline. Keywords—medical image; elastic registration; feature extraction; B-spline