Accelerated 3D-OSEM image reconstruction using a Beowulf PC cluster for pinhole SPECT

Area of Science:

• Medical Imaging
• Computational Science
• Computer Engineering

Background:

• Conventional pinhole SPECT imaging suffers from non-uniform spatial resolution and axial blurring.
• Three-dimensional (3D) imaging with improved resolution is achievable using two-orbit data and 3D Ordered Subsets Expectation Maximization (OSEM).
• The iterative nature of 3D-OSEM and doubled data size from two orbits lead to lengthy reconstruction times.

Purpose of the Study:

• To reduce the long computation time associated with two-orbit pinhole 3D-OSEM reconstruction.
• To implement parallelization of the 3D-OSEM reconstruction process using a Beowulf personal computer (PC) cluster.

Main Methods:

• A Beowulf PC cluster with seven PCs connected via Gbit Ethernet was used.
• Message Passing Interface (MPI) protocol facilitated the parallelization of the reconstruction process.
• Projection data subsets were distributed across PCs, with partial image data shared for updating the overall image estimate.

Main Results:

• Parallelization significantly decreased reconstruction time; a 54-minute single-PC process reduced to 10 minutes with six or seven PCs.
• A speed-up factor of 5.4 was achieved.
• Reconstructed images from the PC cluster were virtually identical to those from a single PC, maintaining image quality.

Conclusions:

• Parallelization of 3D-OSEM reconstruction for pinhole SPECT using a PC cluster effectively reduces computation time.
• The implementation is straightforward and cost-effective.
• This approach offers a practical solution for accelerating high-quality 3D SPECT image reconstruction.