Medical Image Registration Based On Colony Intelligent Algorithm

Medical image registration is one of the important techniques in the field of medical image processing. It also takes a more and more important role in clinical diagnosis and therapy. Although lots of researches on medical image rigid registration have been made for many years, there are still some drawbacks within the current main methods, which have to be corrected further and better for the clinical applications.Now, the proposed approaches of image registration are either based on grey correlation or based on features. There are some advantages and disadvantages on the above mentioned two methods. Following the image registration process based on features, which absorbs ideas of mutual information(MI), the paper further discusses feature detection, optimization algorithms and similarity estimation. In conclusion, this paper is made of several components following:Firstly, the paper compares several traditional edge detection algorithm and proposes an edge detection measure based on mathematical morphology .The conventional methods of edge detection have good effect on clear edge, but are sensitive to noise. Though Mallat used multiresolution wavelet to detect edge of noisy images, it was not effective for much noisy image. Multiresolution wavelet is widely used in antinoise application because of the correlation of different layer wavelet coefficients. This paper presents a new edge detect algorithm based on wavelet multiscale product to suppress noise. The new method is robust on noisy edge and feature points detection.Secondly, the local optimization and global intelligent optimization algorithms are discussed. Due to the lower local search ability and the lack of higher diversity of particles in QPSO, an hybrid particle swarm optimization was proposed, where the Powell optimization is combined to QPSO to enhance the ability of local search. Compared with GA, PSO and QPSO, the performance of PQPSO was demonstrated through MR-CT registration .Thirdly, combining feature pionts with mutual information, a new impoved method for image registration is proposed based on mutual information of feature points. The first step is to detected edge image and feature points by using wavelet multi-scale product.In this step, wavelet multi-scale product can ensure the accuracy of edge image and feature points. Only the mutual informations of pairs of feature points are need to compute, so its computation is very low.Compared with feature points,image edge (line)covers more image information,which is beneficial to the robusticity of registration .According to the theory of mutual information, a new image registration method based on mutual information about distance field of image edges was presented. The distance field of the reference image edges and the binary edge image of floating image were regarded as two discrete probability distributions, and the mutual information between them was adopted as similarity function to register images. Results of experiments show that the new method is excellent to both identical and un-identical image edges.