An Improved Powell Method for Registration of MRI and MRA

摘要

The medical image registration method based on mutual information (MI) has been widely used recently, especially in multimbdality image registration. However, it induces misregistration in monomodality image registration more readily, which is due to the interpolation artifact giving rise to local maxima and sometimes a biased global maximum. Combination of global optimization method (e.g. genetic algorithm) and local optimization methods (e.g. powell method) has been adopted to deal with the problem. In this paper, image registration using brain MRA and MRI dataset is studied. An improvement only based on local optimization method is proposed to overcome impact of interpolation artifact in monomodality image registration. Some constraints, which are directed at parameters of rigid body transformation, are introduced in Powell method so as to avoid getting into local extrema by adjusting searching direction. The registration results based on a MRA dataset and its transformation with known parameters, and the registration results based on MRA and MRI T2-weighted datasets of 20 patients indicate that the proposed algorithm is able to efficiently converge at global optimization vector. Its accuracy and robustness make the algorithm very well suited for applications.