Quantitative performance of advanced resolution recovery strategies on SPECT images: evaluation with use of digital phantom models.

ABSTRACT

Several resolution recovery (RR) methods have been developed. This study was aimed to validate the following performance of the advanced RR methods: Evolution, Astonish, Flash3D, and 3D-OSEM. We compared the advanced RR method with filtered back projection (FBP) and standard order-subset expectation maximization (OSEM) using resolution (RES), cylinder/sphere (CYS), and myocardial (MYD) digital phantoms. The RES phantom was placed in three spheres. Sixteen spheres (hot and cold) were then placed in a concentric configuration (diameter 96-9.6 mm) inside the CYS phantom. The MYD phantom was created by computer simulation with the use of an electron γ-shower 4 (EGS4) and it included two left ventricular defects in the myocardium. The performance was evaluated at source-to-detector distances (R-distance) of 166, 200, and 250 mm with reconstruction parameters (product of subset and iteration SI) with use of the resolution recovery factor, count recovery, normalized mean square error (NMSE), and %CV. According to increased SI updates, the value of the FWHM decreased, and the effect was more obvious as the R-distance increased. The spatial resolution of the advanced RR method was 20 % better than that of FBP and OSEM. The resolution recovery ratio was 80 %, and the count recovery was maintained only in objects with a diameter of >30 mm in the advanced RR method. The NMSE and %CV was 50 and 30 % improved over FBP and OSEM, respectively. The advanced RR method caused overestimation due to Gibbs's phenomenon in the marginal region when the diameter of the sphere was 16-28.8 mm.