Evaluation of a stochastic approach to 2D-3D intensity-based registration

this paper assesses a new 2D-3D rigid registration method based on stochastic clustering and an enhanced estimator for mutual information. It combines precision, accuracy and acceptable computation time. Spine datasets (fluoroscopy and computed tomography) with their gold standard transformations are used. Both optimization method and similarity measure are assessed separately using standardized evaluation methodologies. Sub-millimeter accuracy and the high convergence rate obtained within one minute are compared to other quasi-global optimization processes such as particle filtering.