Parallel constructs for three-dimensional volume registration on an SIMD processor

This paper reports on a parallel algorithm for registration of 3D volumetric images on a single-instruction multiple-data-stream computer. A 3D volume registration algorithm is being implemented on a massively parallel computer (16,384 processor Connection Machine-2). A 3D MR brain image set is used to correlate anatomic information with the physiologic activity from a 3D PET image. The maximum number of stochastic sign changes (SSC) in the scanned subtraction volume is used as the criteria for optimizing the image registration. Presently, this technique is very slow on conventional computers (60 minutes)