Bayesian reconstruction for SPECT: validation with Monte Carlo simulation, experimental phantom, and real patient data
A method for Bayesian image reconstruction from projections is applied to Monte Carlo simulation, experimental phantom, and real patient data from a SPECT acquisition system. This statistical image reconstruction method has three distinct aspects: it uses a priori information about image density distribution of a multinomial process; it considers a spatial correlation of nearby image elements; and it incorporates the Poisson nature of photon detection fluctuation. The Monte Carlo simulation data are generated by computer codes for selected mathematical phantoms containing hot and cold rods. The experimental phantom data are acquired with a Triad SPECT system using radioactive phantoms containing hot and cold rods. The real patient data are obtained from a patient brain scan using the Triad SPECT system. A parallel beam geometry is used. The data are acquired from 120 projection angles uniformly distributed from 0 to 360 degrees. At each projection angle, a 128×128 projection image is measured. This 128×128 projection samples are equally spaced along the axis of detector rotation and perpendicular to the axis, respectively. Each image slice is reconstructed using a 128×128 pixel array. Comparisons between this Bayesian method and maximum likelihood method and filtered backprojection method are given. An improvement in noise suppression is demonstrated using the Bayesian method while image resolution is preserved
Duke Scholars
Published In
Volume
Issue
Start / End Page
Related Subject Headings
- Artificial Intelligence & Image Processing
- 0801 Artificial Intelligence and Image Processing
Citation
Published In
Volume
Issue
Start / End Page
Related Subject Headings
- Artificial Intelligence & Image Processing
- 0801 Artificial Intelligence and Image Processing