Can Artificial Intelligence Accurately Detect Urinary Stones? A Systematic Review

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Urology

Background:

• Urinary stone detection and measurement are crucial in clinical practice.
• Artificial intelligence (AI) shows promise in automating medical image analysis.
• Systematic reviews are needed to evaluate AI's current performance in urinary stone detection.

Purpose of the Study:

• To systematically review the performance of artificial intelligence (AI) in detecting urinary stones using noncontrast computed tomography (NCCT).
• To assess the methodologies and reporting quality of studies evaluating AI for urinary stone detection.

Main Methods:

• A systematic search was conducted across Scopus, Web of Science, Embase, and PubMed.
• Studies were included if they investigated AI for stone detection or measurement on NCCT.
• Risk-of-bias was assessed using established tools, including the Checklist for Artificial Intelligence in Medical Imaging (CLAIM).

Main Results:

• Twelve studies were included, primarily retrospective and single-center.
• AI models demonstrated high performance in stone detection, with sensitivity, specificity, and accuracy ranging from 58.7% to 100%.
• Automated stone volume measurement showed high correlation with manual measurements (r=0.95).

Conclusions:

• AI algorithms can effectively automate urinary stone detection from NCCT.
• Future studies should focus on differentiating stones from phleboliths, external validation, and including complex cases like anatomical abnormalities and urologic foreign bodies.