Research Information System
NUCLEAR MEDICINE COMMUNICATIONS, vol.47, no.2, pp.160-167, 2026 (SCI-Expanded, Scopus)
Background Accurate lesion volume estimation is essential for reliable voxel-based dosimetry in Lu-177 radionuclide therapy. Conventional fixed-threshold segmentation-particularly the commonly used 40% threshold-can markedly underestimate small lesions due to partial volume effects, leading to substantial errors in quantitative SPECT-based dosimetry. Purpose This study systematically evaluated the relationship between lesion size and optimal threshold values in Lu-177 SPECT/CT imaging, quantified deviations introduced by the fixed 40% threshold, and established size-specific adaptive thresholds to improve segmentation and activity recovery accuracy. Methods A NEMA IEC body phantom with six spherical inserts (0.52-26.5 cm(3)) was filled with 20 mCi (740 MBq) Lu-177 at an 8 : 1 lesion-to-background ratio. SPECT/CT data were acquired using 60-90 projections with 10-20 s per frame. Images were reconstructed under 180 parameter combinations varying iterations, subsets, and filters. For each sphere, segmentation was performed using the fixed 40% threshold (40%ThS) and an adaptive, volume-matched threshold (AV%ThS) that reproduced the true physical volume. Results Optimal thresholds showed a strong inverse correlation with lesion size, decreasing from similar to 83% (1.15 cm(3)) to similar to 42% (26.5 cm(3)). The fixed 40% threshold substantially underestimated volumes less than 25 cm(3), with quantitative deviations reaching 45% compared to AV%ThS. Best quantitative recovery was achieved with 90 projections x 20 s and OSEM 10 x 10 iterations/subsets with Butterworth filtering (0.45 cycles/cm, order 10). Conclusion A single fixed threshold is insufficient for accurate Lu-177 SPECT/CT dosimetry across diverse lesion sizes. Size-adaptive thresholding combined with optimized reconstruction parameters improves lesion delineation, enhances quantitative accuracy, and reduces dosimetric uncertainty in clinical practice.