Abstract
This study characterized the recovery-variability trade-off, count-level dependence, and region-specific behavior of open-source block-sequential regularized expectation maximization (BSREM) reconstruction for 123I-FP-CIT single-photon emission computed tomography acquired on a conventional Anger-type SPECT/CT system. Striatal phantom data with normal and asymmetric reduced-putaminal uptake were reconstructed using literature-based ordered-subsets expectation maximization (OSEM) implemented in the vendor software, PyTomography OSEM, and PyTomography BSREM. A BSREM operating point (β = 0.08, γ = 6.0) was selected using background coefficient of variation (CV) and whole-striatal uptake ratio (SUR). Reduced-count datasets at 75%, 50%, and 25% of the original counts were generated by binomial thinning. Metrics included CV, regional SUR, asymmetry index, caudate-to-putamen ratio (CPR), and full width at half maximum (FWHM). In the normal phantom, BSREM yielded lower CV and higher SUR than both OSEM implementations. In the abnormal phantom, BSREM generally yielded higher regional SURs, whereas mean caudate FWHM was 22.6 mm for vendor-software OSEM and 16.9 mm for PyTomography OSEM, and 17.5 mm for BSREM. However, reduced left-putaminal uptake remained underestimated, and putaminal asymmetry moved further from their theoretical values. At the 25% count level, background CV increased to 30.0 ± 6.1% with BSREM and 28.0 ± 5.9% with PyTomography OSEM. Open-source BSREM therefore showed task-, region-, and count-dependent behavior, supporting task- and count-specific parameter selection before clinical application.