ABSTRACT
A major drawback of 3-D medical image registration techniques is the performance bottleneck associated with similarity calculation. Such bottlenecks limit registration applications in clinical situations where fast execution times are required and become particularly apparent in the case of volumetric data sets. In this chapter, a framework for high-performance intensity-based volume registration is presented. Geometric alignment of both reference and sensed volumes is achieved through a combination of translation, rotation, and similarity evaluation. Crucially, similarity estimation is performed intelligently
by agents. The agents work in parallel and communicate with one another by means of a distributed blackboard architecture. Partitioning of the blackboard is used to balance communication and processing workloads. The framework described demonstrates the flexibility of coarse-grained parallelism and shows how high-performance registration can be achieved with nonspecialized architectures.
