A research laboratory proves that choosing the right computing environment can significantly improve performance
UltraSPARC-based systems from Sun Microsystems are helping the Medical Imaging Research Laboratory expand its research horizons and, at the same time, improve the delivery of radiological services to healthcare providers in the Salt Lake City area. The laboratory is the research arm for image processing and computing in the Department of Radiology at the University of Utah's School of Medicine. It also provides a central computing resource composed entirely of Sun servers and workstations for nuclear medicine and diagnostic radiology activities at University Hospital, the local Veteran's Administration Hospital and a private managed care company.
Recently, the laboratory added a four-processor Sun HPC 5000 server, a four-processor Sun HPC 3000 server and a SPARCstorage Array 214 RSM disk subsystem with 80Gb of disk to its SPARCcenter 2000 and two SPARCserver 1000E servers to create a core environment for heavy-duty computing tasks. As Don Baune, the department's director of information technology explains: "We switched to Sun systems in 1988 from DEC equipment because we got a lot better price/performance with the Sun systems. It was much more cost effective at that time and it's stayed that way with the Ultra-SPARC-based systems." In addition to enlarging its server capabilities, the laboratory also added 18 Sun HPC 1 and two Sun HPC 2 workstations as desktop systems supporting its research. Image processing programs are written in C or Fortran by the laboratory, but packages such as Analyze, from the Mayo Clinic, are used for most of the image visualisation steps. It also uses Medx and IDL pseudolanguages for image manipulation.
Baune praises the UltraSPARC systems as excellent, high-performance systems for his work because of Sun's balanced system design - there is consistent performance among all system elements, from processor speed to backplane bandwidth and I/O capabilities. This balance is particularly important when handling medical images that are typically 4Mb in size for a single chest X-ray, a series of Nuclear Magnetic Resonance studies, or a Computerised Tomography series, and even larger for a complete nuclear medicine study. "Other vendors have fast chips that look good at first glance, but when you work on problems that exceed their cache size, the performance really drops off. That's not true with the Sun UltraSPARC systems, they sustain their performance and do very well on these multi-megabyte images."
With the added power and versatility of the HPC 3000 server, the laboratory can do things now that were all but impossible before. Researchers can now extend simulations of nuclear medicine scans to greater size and complexity as they investigate techniques to achieve faster processing times and higher image quality. For example, the system allows researchers to develop and test image-reconstruction algorithms for radiological image sets as large as 1024 x 1024 x 256 bytes, an eight-fold increase in earlier capabilities.
The HPC 3000 also allows the laboratory to provide real-time image-processing support to nuclear medicine units at University Hospital, the local VA hospital, and a managed care company for the first time. Image transport, analysis and distribution used to take several hours - a delay which sometimes led to repeating the whole radiological procedure - a complex, costly and physically stressful process. Now scintillation cameras in the radioisotope scanners send digital data directly to the HPC 3000 for primary image interpretation. After interpretation, physicians at the three medical centres look at the volume visualisation images on a SPARCstation 20 or SPARCstation LX workstation to evaluate their scans' quality and suitability before releasing the patients. Any necessary further scanning is done immediately, a relief to all concerned.
Splitting these numeric-intensive computations across the HPC 3000's multiple processors is the key says Baune. "Many of our problems can divide into smaller units for processing. The Sun compilers multi-thread our code very nicely on the first pass. We tweak it just a little and we get great performance when we run those programs on the HPC 3000. It scales almost linearly with the number of processors, which makes it very affordable and gives us a built-in redundancy."
The laboratory also uses the HPC 3000 in other areas that are not so compute-intensive. The server distributes image files from a nuclear medicine database from Picker International, and it hosts the ShowMe videoconferencing application that the Radiology Department uses to teach physicians in the radiology residency program at multiple sites. Future plans include using Java technology for image distribution. Baune says, "The UltraSPARC systems make our life a lot easier. Overall, they're great."
Compiled by Ajith Ram
(c) 1994-1999 Sun Microsystems, Inc