Beyond a negative PSMA PET: artificial intelligence for occult pN1 risk and extended pelvic lymph-node dissection selection in prostate cancer—a systematic scoping review

A negative prostate-specific membrane antigen (PSMA) PET scan (miN0) does not exclude microscopic pelvic nodal metastasis. Artificial intelligence (AI), machine learning and radiomics might refine residual pathological node-positive (pN1) risk and selection for extended pelvic lymph-node dissection (ePLND). We conducted a PRISMA-ScR systematic scoping review of PubMed/MEDLINE and Scopus (1 January 2018-19 August 2026), supplemented by OpenAlex sensitivity searching and backward citation screening. After deduplication, 496 database records were screened; 99 full reports were assessed and 79 AI/radiomics reports included: 29 nodal-prediction, 22 PET detection/surrogate-reference and 28 prognostic reports. A separate conventional-model search retained 55 PSMA/MRI-aware comparators. No AI study provided direct decision evidence in a dedicated miN0 cohort. Four PSMA-PET AI studies used patient-level surgical pN1, but none was miN0-specific; the only explicit external PSMA-radiomics validation fell from an AUC of approximately 0.88 to 0.57. Overall risk of bias was high in 25/29 nodal reports, driven mainly by Analysis (25/29 high) and Participants (23/29 high). Threshold consequences and absolute pN1 misses were rarely reported. Current AI evidence does not support omitting ePLND after a negative PSMA PET. Dedicated miN0 validation, calibrated probabilities, prespecified thresholds and prospective decision-impact assessment are required.