AI in Enhancing Precision Medicine for Urological Cancer: A Meta-analysis
DOI:
https://doi.org/10.64149/J.Carcinog.24.10s.101-121Keywords:
Artificial intelligence; Machine learning; Deep learning; Precision medicine; Urological cancer; Prostate cancer; Bladder cancer; Kidney cancer; Diagnosis; Prognosis; Meta-analysisAbstract
Background: Artificial intelligence (AI) has emerged as a transformative technology in precision medicine for urological cancers. This meta-analysis systematically evaluates the performance, clinical utility, and implementation challenges of AI applications across the urological oncology care continuum.
Methods: Following PRISMA guidelines, we searched multiple electronic databases from January 2015 to July 2025, identifying studies evaluating AI in urological cancers. We assessed diagnostic accuracy, treatment planning capabilities, and prognostic performance using bivariate random-effects models. Quality assessment was performed using modified QUADAS-2 and AI-specific extensions (STARD-AI, TRIPOD-AI).
Results: We included 142 studies (78 prostate, 38 bladder, 24 kidney, 2 testicular/penile cancer). In diagnostics, AI demonstrated high performance in detecting clinically significant prostate cancer on mpMRI (AUC: 0.93, 95% CI: 0.91-0.95), automated Gleason grading (κ: 0.86, 95% CI: 0.83-0.89), and differentiating renal masses (sensitivity: 0.91, specificity: 0.88). For treatment planning, AI systems reduced radiation planning time by 62% while maintaining plan quality. Prognostically, AI models outperformed conventional tools for biochemical recurrence prediction (C-index: 0.81 vs. 0.73 for CAPRA, P<0.001) and recurrence in bladder cancer (C-index: 0.78 vs. 0.69 for EORTC, P<0.001). However, only 12.7% of studies reported prospective validation, 4.9% documented clinical workflow integration, and 26.8% evaluated algorithmic bias.
Conclusions: AI demonstrates superior performance over conventional approaches in diagnosing, prognosticating, and planning treatment for urological cancers. However, significant gaps exist between algorithmic development and clinical implementation. Future research should prioritize prospective multicenter validation, interpretability, bias assessment, and evaluation of patient-centered outcomes to realize the full potential of AI in precision urological oncology..
