Hybrid optimization for ultrasound and multimodal image registration

Optimization of similarity metrics is an important component of many biomedical image registration applications that greatly affects registration outcome. Selection of an optimization approach is dependent on the modalities being registered, the characteristics of the similarity function, and available computational resources. This paper addresses optimization approaches for 2D and 3D rigid body intra- and multimodal ultrasound registration. Stochastic and direct techniques are compared for mutual information and correlation ratio functions. A new direct/stochastic hybrid approach based on the tabu search is also proposed. Visualization and experimental results suggest the usefulness of such an approach. Results also show that the tabu hybrid technique compares favorably with traditional techniques, based on accuracy and number of function evaluations.