An Automated Image Registration Methodology Using a Hybrid Genetic Algorithm Strategy

W. Huang, J. M. Sullivan Mechanical Engineering Dept., Worcester Polytechnic Institute, Worcester, MA, United States Introduction Aligning multiple medical image volume sets for a composite fMRI analysis is a time consuming process. [1] Conventional gradient-search registration strategies are foiled frequently due to the presence of numerous local minima. In this article an automatic registration algorithm based upon a genetic algorithm (GA) is presented. It eliminates the local minima/maxima traps of conventional optimization techniques. A preprocessing stage normalizes the intensities of each volume set and the images intensity centroids are automatically translated to the origin prior to application of the GA. This strategy facilitates the automated application of masks. It increases the speed and accuracy of registration. The method was applied to fMRI data for Sprague-Dawley rat brains.