Masters Theses

Date of Award

12-1993

Degree Type

Thesis

Degree Name

Master of Science

Major

Computer Science

Major Professor

David Straight

Committee Members

Brad Vander Zanden, Gary Smith, Jack Dongarra

Abstract

In this thesis, I integrated Parallel Virtual Machine (PVM) software (University of Tennessee and Oak Ridge National Laboratory), a state space matrix analysis package, (CtrIC, Integrated Systems, Inc., Santa Clara, OA) and an existing computer workstation network to reduce calculation times for myocardial blood flow (MBF) parametric images. Dynamic heart scanning is one of the most frequently utilized clinical procedures in Positron Emission Tomography (PET). A current drawback for physicians is a lack of quantitative data on myocardial blood flow rates for each patient. There is no baseline, no database of parametric MBF images, and no histories that can be compared and studied. Several mathematical models have been developed to quantify MBF for every pixel on an image. A two compartment model, utilized at the University of Tennessee Medical Center at Knoxville, was implemented in CtrIC, a computer-aided engineering workbench tool which provides state space matrix analysis. This model is computationally intensive and requires significant computer processing time when implemented. Using a single CPU approach with a Sun Sparc 2 processor, each MBF parametric image pixel calculation requires 3.25 sec using CtrIC on a Sun Sparc 2, and a typical MBF parametric image calculation requires nearly 7 hrs. To address this, I chose to parallelize the problem on an existing Sun Workstation network with PVM. PVM executes slave programs running CtrIC software simultaneously across multiple networked processors to create a multiprocessor virtual machine (VM). We reduced processing time from 7 hrs to 1 hr by using our five most powerful workstations with an equivalent power of 4.7 Sun Sparc 10/30 workstations. Data transfer across the network (PVM overhead) and data management contributed less than 3% of the total runtime for this VM configuration. Since the CtrIC integration involved a test product, a profile of the application was performed to determine the speedup anticipated with various product enhancements to this test product. In conclusion, integration of PVM with compartmental model analysis applications (such as myocardial blood flow) offers several benefits; (1) available processors on the network provide computational power needed to calculate parametric images, (2) the compartmental model (in CtrIC) may be modified without recompiling the application code, and (3) parametric image construction times can be estimated given the components of the VM. Key words: Positron Emission Tomography, Parallel Virtual Machine, compartmental modeling, parametric imaging, myocardial blood flow

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS