Modern imaging technologies are capable of producing three- dimensional scenes for use in scientific, industrial and medical applications, thus presenting the need for image processing and image analysis algorithms designed for manipulating 3-D volume data sets. Three-dimensional images in general, and biomedical images in particular, present large scale computationally- intensive image analysis problems especially amenable to parallel processing solutions. In order to fully exploit high performance/supercomputing technologies and move biological image processing to a parallel environment, two key research areas must be investigated. The first is intelligent algorithm design for the most efficient means of programming parallel processors. Parallel computing versus sequential computing algorithms for 3-D biomedical image processing will be studied utilizing a Connection Machine 2 (CM-2) 16k fine-grain single-instruction multiple-data stream (SIMD) supercomputing node, a Stardent GS- 2025 four-processor vector computer and a conventional serial SUN Microsystems reduced instruction set computer (RISC). The second topic to be studied is the interprocessor communication bottleneck in large scale image processing. The use of data structures and programming methods appropriate for efficient manipulation of large data sets will be investigated. The algorithm design will be generic, modular and flexible so that the software can be easily modified as needed with hardware technology improvements. This can reduce the cost of high performance processing and assist in bringing massively parallel computing into the general scientific community for use as a research and teaching tool with broader applicability. Results of this investigation will be applied to biomedical images.
