Beyond the Tumor: Recurrence-Prone Radiomics for Prognostication in Negative Psma Pet/Ct Scans of Prostate Cancer Publisher Pubmed



Yousefirizi F ; Harsini S ; Mohebi M ; Alberts I ; Yusufaly T ; Luo M ; Abdollahi H ; Yazdani E ; Mirabedin S ; Sabouri M ; Geramifar P ; Sheikhzadeh P ; Martineau P ; Wilson D Show All Authors
Source: Biomedical Physics and Engineering Express Published:2026

Abstract

Purpose. Prostate-specific membrane antigen (PSMA) PET/CT is routinely used to restage prostate cancer (PCa) in patients with biochemical recurrence (BCR), yet negative scans may still harbor subclinical disease. This study investigated whether radiomics features extracted from recurrence-prone organs on negative [18F]DCFPyL PET/CT can predict clinical progression (CP) and clinical progression-free survival. Materials and Methods. We performed a post-hoc analysis of 132 patients with BCR (mean age, 74.5 years) who had negative [18F]DCFPyL PET/CT scans and received no further treatment after imaging recurrence-prone organs, defined as anatomical sites at highest risk for PCa recurrence (prostate bed, lymph nodes, bones, liver, and lungs), were segmented using nnU-Net (TotalSegmentator), and radiomics features were extracted. Machine learning models (XGBoost) and Cox proportional hazards models were evaluated using nested cross-validation (5-fold outer, 3-fold inner). External validation across two independent centers was performed after ComBat harmonization. The effect of reader-assigned diagnostic certainty on predictive performance was also examined. Only recurrence-prone organs were analyzed, as prior studies have shown that more than 80% of PCa recurrences arise in these regions. Results. Median PSA at imaging was 0.74 ng ml−1. During a median follow-up of 25.5 months, 42 of 132 patients (31.8%) developed progression. The concordance index improved from 0.65 using clinical variables alone to 0.74 when PET, CT, and clinical features were combined (p < 0.05). Predictive performance was strongly influenced by diagnostic certainty, with moderate-certainty negative scans showing higher accuracy than high-certainty negative scans. Radiomics features derived from negative PSMA PET/CT reflected systemic recurrence risk and were associated with early lung metastases. External validation demonstrated less than 10% reduction in performance despite inter-center differences. Conclusion. Radiomics signatures from recurrence-prone organs on negative PSMA PET/CT may reveal subclinical disease and provide complementary biomarkers for risk stratification in BCR patients. © 2026 IOP Publishing Ltd. All rights, including for text and data mining, AI training, and similar technologies, are reserved.