A Study on the Optimal Artificial Intelligence Model for Determination of Urolithiasis

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• Urolithiasis diagnosis relies on accurate stone detection.
• Clinical decision support systems (CDSS) can enhance diagnostic accuracy.
• Artificial intelligence (AI) offers potential for improving stone detection in CDSS.

Purpose of the Study:

• To develop an optimal AI model for detecting ureter stones within a CDSS.
• To compare various AI and image processing techniques for ureter stone detection.
• To evaluate AI models for supporting clinical judgment in urolithiasis diagnosis.

Main Methods:

• Comparison of machine learning (support vector machine) and deep learning (ResNet-50, Fast R-CNN) models.
• Utilized image processing techniques (watershed) for ureter stone detection.
• Evaluated model performance based on sensitivity, true positive (TP), and false negative rates.

Main Results:

• The ResNet-50 deep learning model demonstrated high recognition accuracy with an average sensitivity of 0.93.
• The developed AI platform showed potential for accurate guidance to the stone area during surgery.
• Sensitivity, a measure of TP probability, was a key performance indicator.

Conclusions:

• The ResNet-50 model is effective for ureter stone detection, contributing to improved urolithiasis diagnosis.
• AI models can be customized for specialized urological diseases.
• Future research will expand AI applications in diverse urological conditions.