Author: Krishnendu Saha 👨🔬
Affiliation: Cleveland Clinic Foundation 🌍
Purpose: The goal is to improve quantitative accuracy of a Maximum-a-posteriori expectation maximization (MAPEM) reconstruction of SPECT phantom images by optimizing Gibbs prior parameters.
Methods: A flangeless deluxe Jaszczak phantom with solid spheres of sizes 31.8, 25.4 and 19.1 mm diameter is filled with diluted 20 mCi of Tc-99m in the background and scanned for 128 projection angles in an elliptical orbit for a total 32 million counts in a Siemens T6 SPECT camera. A system matrix defined by the intersection of the projection with pixel is utilized to reconstruct the sphere slice using a in-house developed MAPEM for 20 iterations. The Gibbs prior smoothing (β) and edge preserving (γ) parameters were varied (β=0.001, 0.005, 0.01; γ=1, 10) to find the best compromise of contrast and noise. Quantitative performances were defined by a background recovery coefficient (BRC) from the difference in counts in the background regions of interest (BROI) and sphere ROI by the BROI counts and compared with the noise from BROI standard deviation by BROI counts.
Results:: The BRC is highest for 31.8 mm sphere (45.4%, noise: 16%, β=0.001, γ=10). Increasing smoothing decreased noise substantially (4.1%, β=0.01, γ=1), however BRC reduced to 33%. Increasing γ only increased noise marginally (7.2%) while substantially increasing BRC (43%, β=0.01, γ=10) indicating a good compromise of BRC and noise. This trend is replicated for smaller size spheres (BRC: 31% and 16.7% for β=0.001, γ=1 for 25.4 mm and 19.1 mm sphere respectively, 28% and 12.6% for β=0.01, γ=10).
Conclusion: A MAPEM algorithm with Gibbs prior was applied to demonstrate that smoothing of SPECT images can be achieved with only marginal loss of contrast. This maybe valuable for low count SPECT images where smoothing causes a loss of contrast.