Machine Learning-Based Quantification of Vesicoureteral Reflux with Enhancing Accuracy and Efficiency

2025-03-19

Data and Metadata

AI Summary

To address inter-observer variability arising from subjective grading of vesicoureteral reflux (VUR) in voiding cystourethrography (VCUG), this study proposes an objective, quantitative VUR grading framework. We identify calyceal distortion as a novel, discriminative imaging biomarker for high-grade VUR—previously unreported—and develop six machine learning classifiers (including logistic regression, decision trees, gradient boosting, and neural networks) trained on nine hand-crafted radiomic features extracted from VCUG images. Model evaluation employs leave-one-out cross-validation. All models achieve perfect sensitivity (100%) and specificity (100%), with area-under-the-curve (AUC) values significantly surpassing those of existing clinical grading systems. This fully automated method enables precise, reproducible VUR grading without false positives or false negatives, advancing VUR assessment from qualitative, experience-dependent interpretation toward standardized, quantitative imaging diagnostics. It provides clinicians with a highly reliable and efficient decision-support tool for routine practice.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Segmentation
• Knowledge Representation and Reasoning: Qualitative Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Vesicoureteral reflux (VUR) is traditionally assessed using subjective grading systems, leading to variability in diagnosis. This study explores the potential of machine learning to enhance diagnostic accuracy by analysing voiding cystourethrogram (VCUG) images. The objective is to develop predictive models that provide an objective and consistent approach to VUR classification. A total of 113 VCUG images were reviewed, with experts grading them based on VUR severity. Nine distinct image features were selected to build six predictive models, which were evaluated using 'leave-one-out' cross-validation. The analysis identified renal calyces’ deformation patterns as key indicators of high-grade VUR. The models—Logistic Regression, Tree, Gradient Boosting, Neural Network, and Stochastic Gradient Descent—achieved precise classifications with no false positives or negatives. High sensitivity to subtle patterns characteristic of different VUR grades was confirmed by substantial Area Under the Curve (AUC) values. This study demonstrates that machine learning can address the limitations of subjective VUR assessments, offering a more reliable and standardized grading system. The findings highlight the significance of renal calyces’ deformation as a predictor of severe VUR cases. Future research should focus on refining methodologies, exploring additional image features, and expanding the dataset to enhance model accuracy and clinical applicability.

Problem

• Improving VUR diagnosis accuracy using machine learning
• Reducing subjectivity in VUR grading with image analysis
• Identifying renal calyceal deformation as severe VUR predictor

Innovation

• Machine learning analyzes VCUG images for VUR
• Six predictive models trained with image features
• Renal calyceal deformation predicts severe VUR