Artificial Intelligence (AI)-based tools in the diagnosis and management of prostate cancer: a systematic review and meta-analysis.
Hein Minn Tun, Lin Naing, Owais Ahmed Malik et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis assessed AI tools for prostate cancer diagnosis, prognosis, and management, drawing on studies published from January 2020 to April 2025. Of 43 included studies, most were retrospective and focused on diagnosis; deep learning was the most common model. In random-effects meta-analysis, pooled AUC was 0.845 (95% CI: 0.809,0.881) for diagnostic tools and 0.785 (95% CI: 0.715, 0.856) for prognostic tools. The authors describe performance as promising but note concerns about pooled performance and real-world applicability.
Key findings
- The review included 43 studies, mostly retrospective and diagnostic-focused (n = 29); deep learning was the most common AI model (49%).
- Across 27 diagnostic studies, pooled AUC was 0.845 (95% CI: 0.809,0.881).
- Across 7 prognostic studies, pooled AUC was 0.785 (95% CI: 0.715, 0.856).
- Subgroup pooled AUC was 0.854 (95% CI: 0.808, 0.901) for deep learning and 0.805 (95% CI: 0.756, 0.856) for classical models.
Why it matters
The synthesis suggests that AI tools may support several parts of prostate cancer care, with the strongest evidence base in diagnosis. However, pooled performance estimates do not by themselves establish usefulness in routine clinical settings. The review’s concerns about validation and applicability underscore the need to assess how tools perform across real-world populations and workflows.
What to keep in mind
Most included studies were retrospective and diagnostic-focused. The authors also report concerns about the applicability of tools related to validation methods and about pooled performance.