Evaluation of the maximum-likelihood method for reconstruction images in positron emission tomography

Conference · Wed May 01 04:00:00 EDT 1985 · J. Nucl. Med.; (United States)

OSTI ID:6910878

Mintun, M A; Gorman, J; Swift, A G; Snyder, D L

A maximum-likelihood algorithm (ML) has been previously proposed for the reconstruction of positron emission tomography (PET) images. Herein, the authors compare the relative performance of this new algorithm to the filtered back projectin (FBP) technique for the PETT VI system with measured and simulated phantom data. Using point source data from PETT VI, in low resolution mode, the ML reconstructed image had a resolution which was approximately half of the full-width half-maximum (FWHM) detector resolution. Using PETT VI brain images (o-15 water bolus) ML yielded images with visibly better resolution than FBP. Simulations of data collections on PETT VI were used to quantitate the relationship between image resolution and noise: 1) A pie-phantom containing six uniform wedges and one million events was reconstructed with ML and FBP. The image resolution and standard deviation (SD) of pixel values within the regions of uniform activity were measured. 2) A brain-phantom containing detailed structure, and one million events was simulated for 20 realizations, all of which were reconstructed using ML and FBP. Image resolution and SD of lcm/sup 2/ region values, across the 20 realizations, were measured. In both simulations ML had little to no advantage over FBP for reconstruction of low resolution and low regional SD images. However, if images with higher SD were accepted, the difference in resolution attained between ML and FBP became substantial: FWHM(ML)/FWHM(FBP)approx. =0.70. These results indicate that ML may be preferred over FBP for high resolution reconstruction, approaching, or even less than the detector resolution.

• Research Organization: Washington Univ. School of Medicine, St. Louis, MO
• OSTI ID: 6910878
• Report Number(s): CONF-850611-
• Conference Information: Journal Name: J. Nucl. Med.; (United States) Journal Volume: 26:5
• Country of Publication: United States
• Language: English

Related Subjects

657000* -- Theoretical & Mathematical Physics
71 CLASSICAL AND QUANTUM MECHANICS
GENERAL PHYSICS
ALGORITHMS
BODY
BRAIN
CENTRAL NERVOUS SYSTEM
COMPUTERIZED SIMULATION
COMPUTERIZED TOMOGRAPHY
DIAGNOSTIC TECHNIQUES
EMISSION COMPUTED TOMOGRAPHY
EVALUATION
IMAGE PROCESSING
MATHEMATICAL LOGIC
MATHEMATICS
MAXIMUM-LIKELIHOOD FIT
NERVOUS SYSTEM
NUMERICAL SOLUTION
ORGANS
POSITRON COMPUTED TOMOGRAPHY
PROCESSING
RESOLUTION
SIGNAL-TO-NOISE RATIO
SIMULATION
STATISTICS
TOMOGRAPHY