Harnessing Artificial Intelligence to Predict Spontaneous Stone Passage: Development and Testing of a Machine...

Area of Science:

• Urology
• Artificial Intelligence
• Medical Informatics

Background:

• Ureteral stones are a common condition requiring effective management strategies.
• Predicting spontaneous stone passage (SSP) is crucial for guiding treatment decisions and patient care.
• Current prediction tools may lack sufficient accuracy for optimal clinical utility.

Purpose of the Study:

• To develop and validate artificial intelligence (AI) based calculators for predicting SSP in patients with ureteral stones.
• To compare the performance of AI-derived models against existing prediction tools.

Main Methods:

• Prospective enrollment of patients with solitary ureteral stones (≤10 mm) on CT.
• Development of AI calculators using machine learning (ML) on a training cohort (70% of patients).
• External validation of AI calculators on a separate testing cohort (30% of patients), comparing against the MIMIC tool.

Main Results:

• Fifty-one percent of training patients achieved SSP; smaller and more distal stones were associated with passage.
• Supervised machine learning (SML) calculator achieved an Area Under the Curve (AUC) of 0.737.
• Unsupervised machine learning (USML) calculator achieved an AUC of 0.706, both outperforming the MIMIC tool (AUC 0.588).

Conclusions:

• AI-powered calculators can be effectively developed for predicting SSP in ureteral stones.
• The developed AI models demonstrate superior predictive performance compared to existing tools.
• These AI calculators can enhance clinical decision-making for the management of ureteral stones.