Automated Grading of Vesicoureteral Reflux (VUR) Using a Dual-Stream CNN Model with Deep Supervision.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Pediatric Urology

Background:

• Vesicoureteral reflux (VUR) is a urinary system disorder with potential renal complications, especially in children.
• Accurate grading of VUR via voiding cystourethrography (VCUG) is vital for clinical management.
• Current grading methods can be subjective and time-consuming.

Purpose of the Study:

• To develop and validate a novel multi-head convolutional neural network (CNN) for automated VUR grading.
• To improve the accuracy and efficiency of VUR classification using VCUG images.
• To provide a reliable tool for supporting clinical decision-making in pediatric VUR cases.

Main Methods:

• A dual-stream CNN architecture with a modified ResNet-50 backbone was utilized.
• The model independently analyzed left and right urinary tracts from VCUG images.
• Deep supervision was incorporated to enhance feature learning and detection of subtle VUR patterns.

Main Results:

• The proposed AI model achieved an average area under the receiver operating characteristic curve (AUC) of 0.82.
• Patient-level accuracy for VUR grading reached 0.84.
• The model effectively categorized VUR into no reflux, mild to moderate reflux, and severe reflux classes.

Conclusions:

• The developed multi-head CNN offers a reliable and automated method for grading VUR from VCUG images.
• This AI-driven approach can assist clinicians in making more informed decisions for pediatric patients with VUR.
• The study highlights the potential of deep learning in enhancing the diagnosis and management of pediatric urinary disorders.