Can AI and predictive models accurately predict stone-free status? a systematic review and meta-analysis

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• Artificial intelligence (AI) and predictive modeling, including radiomics, offer potential for improving stone-free status (SFS) estimation after ureteroscopy.
• These AI models aim to integrate clinical, anatomical, and imaging factors for better peroperative decision-making.

Purpose of the Study:

• To assess the current performance of AI and predictive models for SFS after ureteroscopy.
• To identify sources of heterogeneity in existing studies.
• To determine methodological practices for reliable implementation across diverse settings.

Main Methods:

• A systematic search of six bibliographic databases was conducted up to September 19, 2025.
• Studies developing or validating AI/predictive models for SFS post-ureteroscopy were included.
• Independent dual screening, data extraction, and risk-of-bias assessment using QUADAS-AI were performed.

Main Results:

• Five retrospective cohort studies were analyzed, employing various modeling techniques like logistic regression, radiomics, gradient boosting, and ensembles.
• Significant heterogeneity was observed in SFS definitions, imaging modalities (radiography, ultrasound, CT), and follow-up periods.
• Pooled analyses indicated potential associations between stone size and hydronephrosis with SFS, but with wide prediction intervals.

Conclusions:

• AI and predictive models demonstrate acceptable to excellent discrimination for SFS prediction in ureteroscopy.
• Models combining radiomics with clinical/anatomical data showed the highest performance.
• Substantial heterogeneity across studies renders pooled quantitative estimates clinically uninterpretable, hindering widespread reliable implementation.