Abstract
Objective
PSMA tumor volume (PSMA-TV) is increasingly investigated as a quantitative biomarker for treatment assessment and prognostic stratification in metastatic castration-resistant prostate cancer (mCRPC). However, PSMA-TV depends on segmentation methodology and may be affected by reconstruction-related changes in PET image characteristics. This study evaluated the robustness of different PSMA-TV segmentation approaches across two reconstruction presets and investigated systematic bias, individual-patient disagreement, and equivalence between reconstruction settings.Methods
Forty patients with mCRPC undergoing 68Ga-PSMA-11-PET/CT were retrospectively analyzed. PSMA-TV was quantified using three segmentation approaches: a fixed SUV threshold of 4.0 (T1), a liver-referenced adaptive threshold based on physiological liver uptake (T2), and a lesion-specific 41% SUVmax threshold (T3). Two reconstruction presets differing in image reconstruction parameters were compared. Systematic differences were assessed using signed differences, while individual disagreement was evaluated using absolute differences and median absolute percentage differences. Agreement was assessed using intraclass correlation coefficients (ICC) and logarithmic Bland-Altman analysis. Formal equivalence testing was performed using two one-sided tests (TOST) with exploratory margins of ± 10%, ± 20%, and ± 30% of mean PSMA-TV.Results
PSMA-TV measurements demonstrated threshold-dependent reconstruction sensitivity. T3 showed the largest reconstruction-related variability, with a median signed relative difference of - 53.9% and markedly increased median absolute percentage difference (53.9%). In contrast, T1 and T2 showed substantially lower proportional disagreement (12.7% and 12.1%, respectively). T2 demonstrated the smallest systematic bias between reconstructions (median signed difference 0.4 mL; relative difference 1.1%) compared with T1 (16.9 mL; 9.0%). ICC values indicated high consistency for T1 (0.96; 95% CI, 0.93-0.98) and T2 (0.97; 95% CI, 0.94-0.98), whereas T3 showed lower precision (0.82; 95% CI, 0.29-0.93). Log-ratio Bland-Altman analysis demonstrated no evidence of proportional bias for any threshold. T2 was the only approach demonstrating equivalence within the strictest exploratory ± 10% margin.Conclusions
Reconstruction-dependent effects on PSMA-TV are strongly influenced by segmentation methodology. Liver-referenced adaptive thresholding minimized systematic reconstruction-related bias and demonstrated the highest robustness among evaluated approaches, although individual variability was not completely eliminated. Relative SUVmax-based segmentation showed substantially greater susceptibility to reconstruction changes. These findings support further evaluation of liver-referenced thresholding for standardized PSMA-TV assessment in longitudinal and multicenter settings.