DISTRIBUTED 3-D ITERATIVE RECONSTRUCTION FOR QUANTITATIVE SPECT

We describe a distributed three dimensional (3-D) iterative reconstruc tion library for quantitative single-photon emission computed tomograp hy (SPECT). This library includes 3-D projector-backprojector pairs (P BPs) and distributed 3-D iterative reconstruction algorithms. The 3-D PBPs accurately and efficiently model various combinations of the imag e degrading factors including attenuation, detector response and scatt er response. These PBPs were validated by comparing projection data co mputed using the projectors with that from direct Monte Carlo (MC) sim ulations. The distributed 3-D iterative algorithms spread the projecti on-backprojection operations for all the projection angles over a hete rogeneous network of single or multi-processor computers to reduce the reconstruction time. Based on a master/slave paradigm, these distribu ted algorithms provide dynamic load balancing and fault tolerance. The distributed algorithms were verified by comparing images reconstructe d using both the distributed and non-distributed algorithms. Computati on times for distributed 3-D reconstructions running on up to 4 identi cal processors were reduced by a factor approximately 80-90% times the number of the processors participating, compared to those for non-dis tributed 3-D reconstructions running on a single processor. When combi ned with faster affordable computers, this library provides an efficie nt means for implementing accurate reconstruction and compensation met hods to improve quality and quantitative accuracy in SPECT images.