Purpose: Reconstruction of striatal dopamine transporter (DAT) SPECT is commonly done by filtered back projection (FBP). We investigated if image reconstruction by 3-dimensional ordered-subset expectation maximization (3D-OSEM) with resolution recovery, which has recently become available for clinical routine, provides a relevant improvement.
Methods: I-FP-CIT SPECT studies of 18 patients with normal to severely decreased DAT binding were reconstructed by FBP, 2D-OSEM (without resolution recovery), and 3D-OSEM, each with 2 different filter settings, yielding 3 data set pairs of relatively low and high resolution and noise: FBP with seventh-order Butterworth filter [cutoff frequency, 0.36 Nyquist (FBPlow) and 0.45 Nyquist (FBPhigh)] and OSEM with 8 iterations and 8 subsets (2D-/3D-OSEMlow) and 6 iterations and 16 subsets (2D-/3D-OSEMhigh), each with 8-mm Gaussian filtering. Mean regional counts, variability of counts (coefficient of variation), and binding potential (BPND) were assessed by volume-of-interest analyses of the caudate nucleus, the putamen, and the occipital cortex (reference region).
Results: On visual inspection, both 2D- and 3D-OSEM-reconstructed images showed an optimal delineation of striatal structures, whereas variability (noise) of nonspecific cortical I-FP-CIT uptake was lowest (most homogenous) with FBPlow, slightly higher with 2D-/3D-OSEMlow, and notably higher for the other methods. Volume-of-interest analyses revealed no significant differences of counts in the occipital reference region in comparison to FBPlow (reference method). In caudate nucleus, counts and, consequently, BPND values increased significantly with FBPhigh (mean BPND change, +5.2%), 2D-OSEMlow/high (+3.7%/+6.2%), and, most notably, 3D-OSEMlow/high (+11.1%/+14.0%). In the putamen, this effect was less pronounced for FBPhigh (+1.8%) and 3D-OSEMlow/high (+5.6%/+6.8%) and failed to reach statistical significance for 2D-OSEMlow/high (-0.2%/+0.8%). Regression analyses indicated excellent correlations of BPND between FBPlow and all other methods (R > 0.97), with the highest regression slopes for 3D-OSEM (1.12-1.16) followed by FBPhigh (1.04-1.06) and then 2D-OSEM (1.01-1.04). The order of the variability of counts in the occipital cortex was as follows: FBPlow (12.5%), 2D-OSEMlow (13.9%), 3D-OSEMlow (14.2%), FBPhigh (15.1%), 2D-OSEMhigh (17.0%), and 3D-OSEMhigh (17.6%).
Conclusions: Three-dimensional OSEM considerably improves DAT SPECT reconstruction by offering an optimal combination of high-resolution delineation of striatal structures, superior recovery of signal and BPND, and sufficiently homogeneous nonspecific tracer uptake of the reference region.