%0 Journal Article
%T Hybrid Particle Swarm Optimization and Its Application to Multimodal 3D Medical Image Registration
%A Chen-Lun Lin
%A Aya Mimori
%A Yen-Wei Chen
%J Computational Intelligence and Neuroscience
%D 2012
%I Hindawi Publishing Corporation
%R 10.1155/2012/561406
%X In the area of medical image analysis, 3D multimodality image registration is an important issue. In the processing of registration, an optimization approach has been applied to estimate the transformation of the reference image and target image. Some local optimization techniques are frequently used, such as the gradient descent method. However, these methods need a good initial value in order to avoid the local resolution. In this paper, we present a new improved global optimization approach named hybrid particle swarm optimization (HPSO) for medical image registration, which includes two concepts of genetic algorithms¡ªsubpopulation and crossover. 1. Introduction In the area of medical image analysis, multimodality 3D image registration is an important issue [1]. The purpose of image registration is to register a target image (moving image) to a reference image (fixed image) so that we can combine the information of two images to obtain more detailed information or some specific features. For example, the PET image usually shows metabolic activity of organs and abnormal tissues clearly but lacks the texture of organ tissues. On the other hand, the MR image is described by much complex intensity to represent the texture of organ tissues well. If we implement the MR-PET image registration to combine the information of two images which are different modality, then we can get the accurate shape, volume, and location of abnormal tissues from the registered image. The registration is a very important and helpful preprocessing technique for medical diagnosis or surgical operations. The processing of registration can be seen as an iterated optimization framework, and it can be divided into 3 parts: transformation, cost function, and optimization. In each iteration, the target image is firstly transformed by transformation according to the parameter of the current time. Then, the reference image and the transformed target image are used to calculate the cost function which can evaluate whether the two images are registered or not under the current parameter of transformation. If the images are not registered, an optimization method will be used to adjust the parameters, and a new iteration will start. The application of registration can be classified with dimensionality of image, modality of image, and model of transformation. There are 2D to 2D, 3D to 2D, and 3D to 3D image registration for many different application. The 3-D to 3-D image registration usually needs to estimate more parameters than the 2-D to 2-D image registration, so it does require a more
%U http://www.hindawi.com/journals/cin/2012/561406/