Case Study: Medical Imaging and Analysis
Medical Imaging and Analysis on multi-core processors
Download the full document as a PDF
Download the Dartmouth Medical School case study as a PDF
High-performance parallel computation can be applied to a large number of applications, but among the most compelling are those in medical imaging. Properly exploiting multi-core processors and accelerators (such as GPUs and the Cell BE) can lead to multiple orders of magnitude of improvement in performance. This has the potential to make significant differences in the cost, quality, and even kinds of health care that can be provided.
For many imaging applications, it is now possible to achieve sufficiently high performance on a single desktop machine that the physical acquisition of the data, not the computation, is the bottleneck. This increase in computational performance means that more sophisticated image reconstruction algorithms can be investigated and used in practice. Better algorithms and implementations can translate directly into health benefits. For example, more powerful reconstruction algorithms for CT may require fewer projections and so expose the patient to less radiation, or may be better at resolving important image details.
Additional computational performance can also be applied to the analysis of images. Dr. Susan Schwarz of Dartmouth College used RapidMind and the GPU in elastography image processing. When her code was implemented the using RapidMind platform to use use parallel computation on a GPU accelerator, it took only 1 second. This was an order of magnitude performance improvement over her C++ implementation and a factor of 240 improvement over her original MATATLAB implementation. This and other performance gains are detailed later in this paper, demonstrating the tremendous advantage provided by the RapidMind Platform in medical imaging.