AI based reclassification of PI-RADS 3 examinations in a biopsy-enriched cohort
Abstract This retrospective single-center study evaluated the association between automated AI categories and biopsy-detected clinically significant prostate cancer (csPCa) in 129 patients with an original PI-RADS 3 assessment. Patients were selected from an institutional biopsy series after excluding active surveillance, with MRI available for retrospective AI analysis. The reference standard was biopsy ISUP grade, and csPCa was defined as grade group ≥ 2; postoperative grading was not incorporated. Biopsy identified csPCa in 66/129 patients (51.2%). AI assigned categories 1–2 to 75 patients (58.1%), category 3 to 9 (7.0%), and categories 4–5 to 45 (34.9%). At AI category ≥ 4, sensitivity was 48.5% (95% CI 36.8–60.3%), specificity 79.4% (67.8–87.5%), positive predictive value 71.1% (56.6–82.3%), and negative predictive value 59.5% (48.8–69.4%). The ordinal AUC was 0.667 (95% CI 0.581–0.753). Among patients assigned categories 1–2, 28/75 (37.3%) had csPCa. In 111 patients with PSA density available, the AUC difference between AI and PSA density was 0.140 (95% CI − 0.021 to 0.302; p = 0.088). Automated categories showed limited discrimination in this biopsy-enriched cohort. These results do not establish safe biopsy avoidance or incremental clinical benefit.
Authors
- Ignacio Rojas
- Alfredo Aliaga (ORCID: https://orcid.org/0000-0003-4578-8949)
- Marcelo A. Orvieto
- Hugo Otaola
- Rodrigo Pinochet
- Renato Souper
- Katerin Retamales
- Pablo Bernier
- Andres Labra
- Ignacio Eltit
Institutions
- Universidad del Desarrollo (CL)
- Clínica Alemana (CL)
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-17
- DOI
- https://doi.org/10.1038/s41598-026-79488-w
- Primary Topic
- Prostate Cancer Diagnosis and Treatment
- Type
- article
- Field-Weighted Citation Impact
- 0.00