An interpretable <sup>1</sup>⁸F-PSMA PET/CT-MRI model for optimizing prostate biopsy decisions.
Authors
Affiliations (10)
Affiliations (10)
- Department of Urology, Beijing Hospital, National Center of Gerontology; National Clinical Research Center for Gerontology; The Key Laboratory of Geriatrics of NHC; Institute of Geriatric Medicine, Chinese Academy of Medical Sciences, Beijing, China.
- Fifth School of Clinical Medicine, Peking University, Beijing, China.
- Department of Nuclear Medicine, State Key Laboratory of Complex Severe and Rare Diseases, Center for Rare Diseases Research, Beijing Key Laboratory of Targeted Radiopharmaceutical Development and Translational nuclear medicine, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
- Department of Nuclear Medicine, Beijing Hospital, National Center of Gerontology; National Clinical Research Center for Gerontology; The Key Laboratory of Geriatrics of NHC; Institute of Geriatric Medicine, Chinese Academy of Medical Sciences, Beijing, China.
- Department of Urology, Peking University First Hospital, Beijing, China.
- Department of Radiology, Beijing Hospital, National Center of Gerontology; National Clinical Research Center for Gerontology; The Key Laboratory of Geriatrics of NHC; Institute of Geriatric Medicine, Chinese Academy of Medical Sciences, Beijing, China.
- Department of Pathology, Beijing Hospital, National Center of Gerontology; National Clinical Research Center for Gerontology; The Key Laboratory of Geriatrics of NHC; Institute of Geriatric Medicine, Chinese Academy of Medical Sciences, Beijing, China.
- Department of Urology, Beijing Hospital, National Center of Gerontology; National Clinical Research Center for Gerontology; The Key Laboratory of Geriatrics of NHC; Institute of Geriatric Medicine, Chinese Academy of Medical Sciences, Beijing, China. [email protected].
- Department of Nuclear Medicine, State Key Laboratory of Complex Severe and Rare Diseases, Center for Rare Diseases Research, Beijing Key Laboratory of Targeted Radiopharmaceutical Development and Translational nuclear medicine, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China. [email protected].
- Department of Urology, Beijing Hospital, National Center of Gerontology; National Clinical Research Center for Gerontology; The Key Laboratory of Geriatrics of NHC; Institute of Geriatric Medicine, Chinese Academy of Medical Sciences, Beijing, China. [email protected].
Abstract
Distinguishing indolent from clinically significant prostate cancer (csPCa) in biopsy-naïve men remains a diagnostic challenge, often leading to unnecessary biopsies. This study aimed to develop and externally validate an interpretable multimodal model integrating <sup>18</sup>F-PSMA PET/CT, mpMRI, and clinical variables to predict csPCa and optimize biopsy decision-making. This multicenter retrospective study analyzed 302 biopsy-naïve patients derived from a prospective multicenter cohort. Radiomic features were extracted from <sup>18</sup>F-PSMA PET and mpMRI. A dual-level region-of-interest (ROI) strategy was used to capture lesion and whole-gland features. To prevent overfitting, the optimal radiomics signature was derived from the intersection of four machine learning (ML) selection algorithms, while clinical features were selected via logistic regression. Eight ML models were trained and tested with ten-fold cross-validation. A Fusion model, combining radiomics and clinical variables, was evaluated against clinical and single-modality models using AUC, decision curve analysis (DCA), and clinical impact curves (CIC). Additionally, SHapley Additive exPlanations (SHAP) values elucidated feature contributions. The Fusion model demonstrated robust diagnostic efficacy, achieving AUCs of 0.92 in both the internal (95% CI: 0.85-0.97) and external (95% CI: 0.83-0.99) validation cohorts. In the PI-RADS 3 subgroup, the model outperformed standard clinical assessments, yielding an excellent AUC of 0.97 (sensitivity 95.0%, specificity 86.5%). Clinically, risk stratification utilizing the Fusion model could avoid 82.9%, 75.0%, and 72.2% of unnecessary biopsies across the three cohorts, while maintaining low missed-diagnosis rates of 4.0%, 9.5%, and 6.1%, respectively. Furthermore, SHAP analysis identified PET-derived intensity and textural features as the dominant predictors, underscoring the indispensable added value of <sup>18</sup>F-PSMA PET. This interpretable Fusion model integrating <sup>18</sup>F-PSMA PET/CT and mpMRI demonstrated superior diagnostic performance for csPCa compared to standard clinical assessment. This approach shows potential as an effective non-invasive triage tool to support biopsy decision-making, with the potential to reduce unnecessary biopsies while maintaining oncological safety.