MRI spin relaxometry is the process of recovering the spin density spectrum from the time samples of the spin signal for each pixel of a magnetic resonance image. Since healthy tissue exhibits different spin relaxation rates from diseased tissue, MRI spin relaxometry potentially has utility for diagnosing disease. However, recovering the spin relaxation rates involves solving an inverse problem which requires substantial computation. The computation's running time can be reduced by processing the pixels in parallel on a parallel computer. A parallel program for solving the MRI spin relaxometry problem, SRSolve, was implemented in Java with MPI, its running time was measured on a 32-processor cluster parallel computer, and its performance was compared to the CONTIN program. CONTIN required about 44 sec on the average to solve one pixel and about 3600 sec to solve an entire 64x64-pixel test image (with 2,597 unmasked pixels) on the parallel computer. SRSolve required 3.04 sec on the average to solve one pixel and 263 sec to solve the entire image on the parallel computer.
Department, Program, or Center
Computer Science (GCCIS)
Alan Kaminsky and Luke McOmber. Solving an MRI spin relaxometry problem with parallel computing. Rochester Institute of Technology Department of Computer Science Technical Report, June 27, 2005.
RIT – Main Campus