Prognostic Value of Baseline PSMA PET/CT Tumor Volume in mCRPC: Comparison of a Semi-Automated and an AI-Based Segmentation Method

Background Radioligand therapy targeting prostate-specific membrane antigen is an established treatment for patients with metastatic castration-resistant prostate cancer. Accurate quantification of tumor burden on PSMA PET/CT is essential for prognostic stratification and therapeutic planning. PSMA-derived tumor volume has emerged as a strong imaging biomarker for overall survival in this population. However, variability in segmentation approaches may affect volumetric measurements and limit clinical comparability. This study compares two segmentation methods—semi-automated threshold-based segmentation using Syngo.via and fully automated, artificial intelligence-driven analysis with aPROMISE—to evaluate their prognostic value for baseline tumor burden in patients undergoing radioligand therapy. Results Baseline PSMA PET/CT scans from 111 patients with metastatic castration-resistant prostate cancer were retrospectively analyzed. Tumor volumes differed significantly between the two segmentation methods, with mean volumes of 190.11 mL using Syngo.via and 1123.63 mL using aPROMISE (p<0.001). Despite this difference in absolute values, measurements were strongly correlated (ρ = 0.865, p < 0.001). Both segmentation approaches demonstrated that higher baseline tumor volume was significantly associated with decreased overall survival. Kaplan-Meier and Cox regression analyses showed similar prognostic performance, with hazard ratios of 3.43 (95% CI: 2.01-5.83) for Syngo.via and 3.70 (95% CI: 2.17-6.32) for aPROMISE. Conclusions Despite notable differences in absolute tumor volume estimates, both Syngo.via and aPROMISE provided comparable prognostic value for overall survival in patients with metastatic castration-resistant prostate cancer undergoing PSMA radioligand therapy. These findings support the integration of tumor volume as a prognostic imaging biomarker in clinical practice. Importantly, consistent use of a single segmentation platform is recommended for reliable longitudinal assessment and clinical decision-making.