Related Experiment Videos
Grouped-coordinate ascent algorithms for penalized-likelihood transmission image reconstruction
J A Fessler1, E P Ficaro, N H Clinthorne
1University of Michigan, Ann Arbor 48109-2122, USA. fessler@umich.edu
IEEE Transactions on Medical Imaging
|April 1, 1997
Summary
New grouped-coordinate ascent (GCA) algorithms improve attenuation map reconstruction from low-count positron emission tomography (PET) scans. These algorithms offer faster convergence and better efficiency than previous methods.
Area of Science:
- Medical Imaging
- Computational Science
Background:
- Accurate attenuation map reconstruction is crucial for quantitative positron emission tomography (PET).
- Low-count transmission scans present challenges for traditional reconstruction algorithms due to noise and computational demands.
Purpose of the Study:
- To introduce a novel class of algorithms for penalized-likelihood reconstruction of attenuation maps.
- To address limitations of existing methods in terms of computational efficiency and constraint handling.
Main Methods:
- Derivation of new algorithms using a convexity technique applied to the transmission log-likelihood.
- Development of grouped-coordinate ascent (GCA) algorithms as a specific implementation.
- Comparison with single-coordinate ascent (SCA) and maximum likelihood-expectation maximization (ML-EM) algorithms.
Main Results:
- GCA algorithms require fewer exponentiations compared to ML-EM and SCA.
- The algorithms inherently handle nonnegativity constraints.
- GCA algorithms demonstrate faster convergence than SCA, even on standard hardware.
- Parallelizability of GCA algorithms offers advantages over SCA.
Conclusions:
- The proposed GCA algorithms represent a significant advancement in attenuation map reconstruction for low-count PET transmission scans.
- These methods offer improved computational efficiency, better constraint accommodation, and faster convergence.
- The findings have implications for enhancing the quality and speed of PET imaging.