FMRI image registration based on normalized maximum mutual information & genetic algorithm

Abstract

On the research of brain and cognitive science based on fMRI, the subject needs to be measured many times during the functional imaging experiments. There is a slight head-movement inevitably during the experiment, so the fMRI images must be registered precisely. An fMRI image registration approach was proposed based on normalized maximum mutual information, Simplex and genetic algorithm. In the algorithm, we utilize normalized maximum mutual information as the similarity measurement; genetic algorithm and Nelder-Mead's simplex method are respectively responsible for the global approximate optimization and local accurate search of image translation and rotation parameters. According to the registration analysis of the time series fMRI images, this algorithm is proved to be a precise, fast approach compared with the traditional direct search method and genetic algorithm. The experimental results show that the registration precision is improved while maintaining robustness.