Abstract
BACKGROUND/PURPOSE: Small-lesion detectability in positron emission tomography-computed tomography (PET/CT) depends on reconstruction parameters such as postreconstruction smoothing and matrix size, but practical guidance for selecting these parameters is limited. This work quantifies how iterative reconstruction settings affect contrast recovery coefficient (CRC, %), background variability (BV), and contrast-to-noise ratio (CNR) using a standardized NEMA IEC body phantom. This phantom-only study quantifies how iterative reconstruction and postreconstruction filtering affect CRC, %, BV, and CNR in a standardized NEMA IEC body phantom. MATERIALS AND METHODS: A NEMA IEC body phantom with six spheres (10, 13, 17, 22, 28, and 37 mm) and a lung insert was filled to achieve nominal 4:1 and 8:1 sphere-to-background activity ratios using ¹3F. Four spheres were prepared as hot spheres and two as cold spheres (4 hot/2 cold). The phantom was scanned on a Philips Gemini PET/CT system. Images were reconstructed using ordered-subsets expectation maximization (3 iterations/16 subsets) while varying (i) Gaussian postreconstruction smoothing kernels of 3 mm, 5 mm, and 7 mm full width at half maximum, (ii) alternative filters (Hann, Butterworth, median), and (iii) matrix size (128 × 128 vs. 168 × 168). For each condition, we measured CRC, BV (expressed as % standard deviation of background VOIs), and CNR. RESULTS: CRC increased with sharper filtering and with the larger matrix. At a 4:1 activity ratio, the 10 mm sphere reached 15% CRC with a 3 mm Gaussian filter, 12% with a 5 mm filter, and 10% with a 7 mm filter; the 37 mm sphere remained around 55% CRC across filters. At an 8:1 ratio, CRC rose overall: the 10 mm sphere reached 17% (3 mm), 14% (5 mm), and 12% (7 mm), while larger spheres maintained CRC ≥60% (for example, the 28 mm sphere reached 69% CRC with 3 mm Gaussian smoothing). Increasing matrix size from 128 × 128 to 168 × 168 improved CRC by approximately 2-7 absolute percentage points depending on sphere size (13 mm: 24% → 28%; 37 mm: 50% → 57%). These CRC gains were accompanied by increased BV and changes in CNR, which are now quantified in this revision. BV and CNR values represent within-scan spatial variability across background VOIs, not variability from repeated acquisitions. CONCLUSIONS: Image quality optimization is a trade-off. A 168 × 168 matrix combined with a 3 mm Gaussian filter maximizes CRC and CNR in the smallest spheres (10-13 mm) but increases BV. A 5 mm Gaussian filter provides a more clinically acceptable CRC-BV balance for spheres ≥17 mm (for example, CRC ≈40-46% for the 17 mm sphere) while moderating BV compared with the sharpest setting. A 7 mm filter suppresses BV further, but at the cost of CRC for the smallest spheres. Reporting CRC alone can exaggerate the benefit of aggressive reconstruction; BV and CNR must be considered together. A 168 × 168 matrix with a 3-5 mm Gaussian improved CRC for ≤13 mm spheres but at a BV penalty; for ≥17 mm spheres, a 5 mm Gaussian provided a more balanced CRC-BV profile, with similar or improved CNR relative to 3 mm. These findings are specific to this NEMA IEC phantom and may not translate directly to patient imaging. All BV and CNR values reflect spatial noise within a single acquisition rather than scan-to-scan variability.