Chronic Kidney Disease Progression

This retrospective study used deep learning to extract anatomical features from initial postnatal kidney ultrasounds to predict chronic kidney disease (CKD) progression in boys with posterior urethral valves (PUV). While clinical models—particularly those driven by nadir creatinine—outperformed imaging-only models, combining imaging features with clinical data improved predictive performance, achieving a C-index of 0.82 at 6 months. These findings suggest that deep learning–derived ultrasound features can enhance early risk stratification when integrated with clinical variables in pediatric PUV patients.