A SIMD architecture for medical imaging

Abstract

This paper addresses parallel processing in 3D medical imaging. A SIMD architecture common for image reconstruction, image processing, and volume visualization is proposed. The ray casting algorithm for volume visualization along with some of its improvements is described. The image-space parallelization of ray casting is given. The parallelization ensures an even load balancing among processors and conflict free memory access without the need of replicating the entire volume data set. This parallelization covers distribution of the volume data set among global memory modules, distribution of the image among local memories, processors - global memory modules mapping, generation and scheduling of image parts, and processor program.