Performance evaluation of block-iterative algorithms for SPECT reconstruction

The purpose of this study is to evaluate the performance of four block-iterative algorithms, ordered-subsets expectation-maximization (OS-EM), rescaled block-iterative EM (RBI-EM), modified row-action maximum likelihood algorithm (RAMLA) and rescaled block-iterative maximum a posteriori EM (RBI-MAP-...

Full description

Saved in:
Bibliographic Details
Main Authors: Chi Liu, Volokh, L., Xide Zhao, Jingyan Xu, Taek-Soo Lee, Tsui, B.M.W.
Format: Conference Proceeding
Language:English
Subjects:
Attenuation
Blood vessels
Collimators
Convergence
Image quality
Image reconstruction
Imaging phantoms
Noise generators
Noise measurement
Scattering
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The purpose of this study is to evaluate the performance of four block-iterative algorithms, ordered-subsets expectation-maximization (OS-EM), rescaled block-iterative EM (RBI-EM), modified row-action maximum likelihood algorithm (RAMLA) and rescaled block-iterative maximum a posteriori EM (RBI-MAP-EM), for In-111 ProstaScint/spl reg/ SPECT image reconstruction. The 3D NCAT phantom with realistic In-111 ProstaScint/spl reg/ activity distribution was used in the study. Noise-free and noisy projections of the phantom obtained using a medium-energy general-purpose (MEGP) collimator were generated using Monte Carlo simulation methods. For each algorithm, the projection data were reconstructed with the compensations for attenuation, collimator-detector response and scatter. Image quality was evaluated in terms of FWHM of a profile through a small blood vessel, normalized mean square error (NMSE), ensemble normalized standard deviation (NSDE) of a uniform region of interest (ROI) in the reconstructed image measured from 30 noise realizations, and regional NSD (NSDR) of an ROI measure from 1 noise realization. The results indicated that, RBI-EM has superior performance than that of OS-EM when less than 4 views per subset were used and similar performance when 4 or more views per subset were used. Modified RAMLA provides similar image quality with a slower convergence rate than that of OS-EM. Using well-chosen parameters, RBI-MAP-EM provides increased noise smoothing with less loss in resolution and error. We conclude that when compared with OS-EM, the RBI-EM and modified RAMLA have the same performance at a slower convergence rate, while the RBI-MAP-EM has superior performance and can potentially improve image quality.
ISSN:1082-3654
2577-0829
DOI:10.1109/NSSMIC.2005.1596676