Predicting Urinary Stone Composition in Single-Use Flexible Ureteroscopic Images with a Convolutional Neural Network

Area of Science:

• Urology
• Artificial Intelligence
• Medical Imaging

Background:

• Urine stone composition analysis is crucial for effective urolithiasis management.
• Single-use flexible ureteroscopes (fURS) offer financial advantages but present lower image quality compared to reusable scopes.
• Predicting stone composition directly from fURS images could enhance treatment strategies.

Purpose of the Study:

• To evaluate the efficacy of a convolutional neural network (CNN) in predicting urinary stone composition using images from single-use fURS.
• To assess the performance of AI in analyzing low-resolution endoscopic images for urolithiasis diagnosis.

Main Methods:

• Retrospective analysis of 207 surgical images from single-use fURS lithotripsy.
• Development of a CNN model using transfer learning with Resnet-18, trained on endoscopic images and stone classification data.
• Classification of stones into Calcium and Non-calcium groups for model training and validation.

Main Results:

• The CNN achieved an overall accuracy of 81.8% in predicting stone composition.
• High performance was observed in the Calcium group (Recall: 88.2%, Precision: 88.2%).
• The model demonstrated a good classification performance with an area under the ROC curve of 0.82.

Conclusions:

• AI, specifically CNNs, can effectively predict urinary stone composition even with low-resolution images from single-use fURS.
• This study is the first to apply AI to single-use fURS images, demonstrating its potential in challenging endoscopic conditions.
• The findings suggest broader applications for AI in urology, potentially improving diagnostic capabilities and treatment outcomes.