10 likes | 136 Views
Multi-modality Image Registration Using Mutual Information Based on Gradient Vector Flow Yujun Guo, Cheng-Chang Lu Department of Computer Science, Kent State University, Kent, Ohio 44242. Introduction
E N D
Multi-modality Image Registration Using Mutual Information Based on Gradient Vector Flow Yujun Guo, Cheng-Chang LuDepartment of Computer Science, Kent State University, Kent, Ohio 44242 Introduction Image registration is important in medical image analysis. Similarity measure plays a critical role in image registration. Mutual information (MI) has been successfully used as a similarity measure for both mono- and multimodality image registration. However, MI measurement only takes statistical intensity information into account, and ignores spatial information in the images. A random reshuffling of the image voxels (identical for both images) will yield the same MI value as for the original images. We propose a novel approach to incorporate spatial information into MI through gradient vector flow (GVF). Multimodality brain image registration is performed to test the accuracy and robustness of the proposed method. • Materials and methods • MI has been proved a successful similarity measure for both mono- and multimodality image registration • MI based on original image intensities has a narrow attraction range • GVF field as external force has a much larger capture range than traditional potential forces • Apply MI to GVF fields instead of the original images may increase the capture range of image gradient Gradient vector flow (GVF) was originally proposed as an external force to attract the snake to the actual boundary of the objects. The vector field v(x,y) = [u(x,y), v(x,y)]is computed by a diffusion process, which can be implemented by minimizing the following energy function: GVF field is found by solving the following two Euler equations: GVF-intensity (GVFI) map of an image I(x,y) is defined as The definitions of edge map f studied in this paper are as follows: Results • Simulated T1 and T2 MR brain images generated by BrainWeb MR simulator are used (http://www.bic.mni.mcgill.ca/brainweb/) • Five image pairs from T1 and T2 volumes at different noise levels • T1 slice is randomly transformed, then is registered to T2 slice, to find the optimal parameters • 300 tests for each method on each pair Fig. 2. Fig. 1. Left: T1 image Right: GVFI Table 1. Success rates of MI-based registration methods where MI is based on original intensity or GVFI corresponding to four types of edge map. References C. Xu and J.L. Prince, Snakes, shapes, and gradient vector flow. IEEE Trans. Image Process. 7(3):359-369, Mar. 1998. F. Maes, A. Collignon et al. Multimodality image registration by maximization of mutual information. IEEE Trans. Medical Imaging, 16(2):187-198, 1997. C. Studholme, D.L.G. Hill and D.J.Hawkes, An overlap invariant entropy measure of 3D medical image alignment. Pattern Recognition, 32(1):71-86, 1999. D.L. Collins et al. Design and construction of a realistic digital brain phantom. IEEE Trans. Medical Imaging, 17(3):463-468, 1998. • Conclusions • The success rate of proposed method is higher than that of traditional MI • MI based on GVFI is robust to noise • GVFI based on f1 performs better than the others