Improvement to global optimization of mutual information on retrospective registration of magnetic resonance images

Robust retrospective 3D image registration is critical for a number of applications, our extensive simulations of mutual information (MI), a robust similarity measure for medical image registration, indicate that local minima in the cost function may caused by various image manipulations. An improvement to global optimization of mutual information for image registration that can be used in conjunction with a multiresolution paradigm is described in this paper, this technique combines differential evolution (DE) algorithm in continuous space with dividing rectangle (DIRECT) algorithm, and is efficient in retrospective registration of magnetic resonance images of brain.