Comparison of 3-D OSEM versus 1-D SAGE for focal total-activity quantification in I-131 SPECT with HE collimation

Abstract

A three-dimensional (3-D) ordered-subset expectation maximization (OSEM) reconstruction algorithm used fits to the average behavior of the I-131 point-source-response function at different depths to model the detector-collimator response of a SPECT camera. The fitting function was a Gaussian, plus a single exponential to account for septal-penetration of the high-energy (HE) collimation. Results from that unregularized 3-D OSEM were compared to those with regularized one-dimensional (1-D) space-alternating generalized expectation maximization (SAGE). The latter used a simple strip integral. Each reconstruction algorithm employed attenuation correction and scatter compensation. The SPECT camera was calibrated by imaging a 200 cm/sup 3/ sphere containing a known I-131 activity. This sphere was located inside an elliptical cylinder. The water in the cylinder contained one of four I-131 activity concentrations. The ratio of background-over-target activity concentration b was used to characterize the different situations. The known activity and the reconstructed counts within the sphere at each b value yielded a counts-to-activity conversion factor, CF. After reconstruction with 3-D OSEM, CF varied much less with b than after reconstruction with 1-D SAGE. A test of activity quantification was carried out with 20 realizations of an anthropomorphic phantom with zero background activity. This phantom simulated a 100 cm/sup 3/ spherical tumor centrally located inferior to the lungs. The mean bias in the estimate of "tumor" activity was -7.3% with 3-D- OSEM-based quantification and -4.3% with 1-D-SAGE-based quantification. Recovery-coefficient-based correction was applied with both algorithms. With UHE collimation, the previously-published mean bias in the estimate of the "tumor" activity was -5.2% after adding a recovery-coefficient-based correction. For that collimation, the previously-published mean bias with 1-D-SAGE-based quantification was -10.7%. We conclude that 1) two-component fitting allowed 3-D OSEM to do a good job of compensating for the septal penetration of the HE collimation, 2) the procedure for activity quantification with 3-D OSEM was simpler than that with 1-D SAGE, and 3) activity accuracy for a test phantom had a similar quality for four situations, using 3-D OSEM-based activity quantification and 1-D-SAGE-based activity quantification and combining each with HE and UHE collimation.