Applications of artificial intelligence algorithms in ultrasound-based kidney stone detection, classification, prediction, and management: a systematic review.
Mohammadreza Elhaie, Abolfazl Koozari, Salman Mohammed Albeshan et al.
✦ AI-curated · Sources linked
In 30 seconds
This PRISMA-guided systematic review assessed artificial intelligence and deep-learning applications for ultrasound-based kidney stone detection, classification, prediction of complications or outcomes, and procedural guidance. Nine original studies were included: three on image detection or segmentation, four on prediction, and two on procedural guidance. Reported performance was high, with accuracy up to 96.54% and prediction AUCs above 0.90; procedural studies also reported improved outcomes. However, heterogeneous datasets and validation approaches prevented meta-analysis, and the review rated the overall evidence as low to very low certainty.
Key findings
- Across included studies, reported diagnostic accuracy reached 96.54%.
- Predictive models reported AUCs greater than 0.90 for complications or outcomes.
- The nine studies covered detection or segmentation (3), prediction (4), and procedural guidance (2).
- Heterogeneity in datasets and validation limited meta-analysis; evidence certainty was low to very low.
Why it matters
Ultrasound is commonly used first for kidney stone assessment, but its sensitivity and operator dependence are limitations. AI approaches may support image interpretation, risk prediction, and procedural guidance across several tasks. The review’s findings are promising, but the small evidence base and low certainty mean reported performance should not be assumed to generalize across settings.
What to keep in mind
Only nine studies were included, with heterogeneous datasets and validation methods that precluded meta-analysis. The review rated the evidence as low to very low certainty.