A machine learning algorithm predicting risk of dilating VUR among infants with hydronephrosis using UTD...

Area of Science:

• Pediatric Urology
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Urinary Tract Dilation (UTD) classification aims for objective grading of antenatal and postnatal UTD.
• Current UTD classifications have unclear associations with anomalies like VUR, leading to subjective management recommendations.

Purpose of the Study:

• To develop a reliable machine learning (ML) model for predicting VUR from early postnatal ultrasounds.
• To improve the objectivity of management decisions for infants with prenatal hydronephrosis.

Main Methods:

• Retrospective review of radiology and medical records for infants (0-90 days) with antenatal UTD.
• Development and validation of an ML model using features like gender, ureteral dilation, and parenchymal appearance.
• Primary outcome was defined as dilating (≥Gr3) VUR, confirmed via VCUG.

Main Results:

• The study included 280 infants (540 renal units), with 66 units diagnosed with ≥ grade 3 VUR.
• The final ML model, incorporating clinical and ultrasound features, achieved an AUC of 0.81 for VUR prediction.
• Key predictors included gender, ureteral dilation, parenchymal appearance, parenchymal thickness, and central calyceal dilation.

Conclusions:

• A machine learning model can effectively predict dilating VUR from early postnatal ultrasounds in infants with hydronephrosis.
• This predictive model supports individualized management strategies for children with prenatal hydronephrosis.
• The model facilitates a more selective and efficient use of voiding cystourethrogram (VCUG).