Two novel deep-learning models to predict spontaneous ureteral calculi passage: Model development and validation

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• Ureteral stones affect a significant number of patients.
• Predicting spontaneous stone passage is crucial for treatment planning.
• Current methods, like stone size, have limitations.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for predicting spontaneous ureteral stone passage (SSP).
• To compare the performance of DL models against conventional methods.

Main Methods:

• Retrospective analysis of 1217 patients with ureteral stones.
• Development of two DL models: 2D ResNet29 and 3D ResNet29.
• Comparison with a conventional approach based on stone diameter (<5 mm).

Main Results:

• The 3D ResNet29 model achieved 90.6% accuracy, 88.2% sensitivity, and 95.1% specificity.
• Both 2D and 3D DL models significantly outperformed the stone diameter approach.
• DL models showed a higher area under the curve in ROC analysis.

Conclusions:

• Deep learning models, especially the 3D ResNet29, are effective for predicting SSP.
• These models can assist clinicians in deciding between surgical intervention and conservative management.
• Novel DL approaches offer improved prediction accuracy for ureteral stone outcomes.