Artificial Intelligence Versus Conventional Methods for NCCN Risk Stratification in Localized Prostate Cancer (2020-2025): A Systematic Review.

BACKGROUND AND OBJECTIVE: Accurate risk stratification in localized prostate cancer is essential for guiding treatment decisions. Conventional National Comprehensive Cancer Network (NCCN) risk groups rely on prostate-specific antigen (PSA), Gleason grade group, and clinical stage, while artificial intelligence (AI) methods, including radiomics, digital pathology, and multimodal prediction models, are increasingly being evaluated as alternatives or complements. This systematic review aims to assess studies published between 2020 and 2025 that directly compare AI-based models with traditional NCCN risk stratification methods in localized prostate cancer. METHODS: A systematic search of PubMed/MEDLINE, Scopus, and Cochrane Library (January 2020-September 2025) was conducted. Eligible studies focused on localized prostate cancer, used AI-based models, compared them with NCCN risk stratification or its components, and reported outcomes such as the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, calibration, or decision-curve analysis. Case reports, nonhuman studies, and abstracts without full data were excluded. Title/abstract and full-text screening was performed independently by two reviewers. Risk of bias was assessed using PROBAST, and reporting transparency was benchmarked against TRIPOD-AI. RESULTS: Out of 686 records, 100 duplicates were removed. Of the 586 screened, 95 were excluded. A total of 491 records mapped to NCCN components and compared AI with conventional methods; 43 full-text studies met inclusion criteria. AI approaches combining MRI radiomics, PET imaging, and histopathology whole-slide analysis consistently showed higher discrimination than NCCN models, especially for predicting adverse pathology and biochemical recurrence. However, calibration, external validation, and reporting quality were addressed inconsistently. CONCLUSIONS: AI-based methods show promise in improving NCCN risk stratification for localized prostate cancer, delivering better prognostic accuracy than traditional approaches. However, variability in methods, limited external validation, and gaps in transparent reporting highlight the need for larger, multiinstitutional prospective studies before these methods can be widely adopted in clinical practice.