Related Experiment Videos
A parallel Monte Carlo code for planar and SPECT imaging: implementation, verification and applications in (131)I
Yuni K Dewaraja1, Michael Ljungberg, Amitava Majumdar
1Department of Radiology, Division of Nuclear Medicine, The University of Michigan Medical Center, 3480 Kresge III, 204 Zina Pitcher Place, Ann Arbor, MI 48109-0552, USA. yuni@umich.edu
Computer Methods and Programs in Biomedicine
|January 26, 2002
Summary
The SIMIND Monte Carlo code was parallelized for faster simulations in Single Photon Emission Computed Tomography (SPECT). This parallel computing approach significantly speeds up complex medical imaging simulations.
Area of Science:
- Medical Imaging
- Computational Physics
- Parallel Computing
Background:
- Monte Carlo simulations are crucial for medical imaging, but computationally intensive.
- Parallel computing offers a potential solution for accelerating these simulations.
- The SIMIND code is widely used for simulating SPECT imaging.
Purpose of the Study:
- To implement and evaluate the SIMIND Monte Carlo code on a distributed memory parallel computer.
- To assess the performance and scalability of the parallelized code for SPECT simulations.
- To validate the accuracy of the parallel code using experimental data.
Main Methods:
- Parallelization of the SIMIND code using the Message Passing Interface (MPI).
- Utilizing the Scalable Parallel Random Number Generator (SPRNG) for uncorrelated random number streams.
- Partitioning photon transport calculations across multiple processors on an IBM SP2 system.
Main Results:
- Achieved a linear increase in computing speed with up to 32 processors.
- Demonstrated significant speed-up for high-energy photon emitters in SPECT simulations.
- Validated code accuracy by comparing simulated and experimental SPECT images for (131)I.
Conclusions:
- The parallelized SIMIND code provides a significant computational speed-up for SPECT simulations.
- This approach is effective for complex simulations, including scatter and attenuation correction.
- The parallelization techniques are transferable to other distributed memory architectures.