Multiparametric Ultrasound and Machine Learning for Predicting Renal Scarring in Children

Background/Objectives: Renal scarring in children is linked to long-term complications, including hypertension and chronic kidney disease. Although dimercaptosuccinic acid (DMSA) scintigraphy is the reference standard, routine use is limited due to radiation exposure. This study evaluated whether multiparametric ultrasound combined with machine learning could predict DMSA-detected renal scarring in pediatric patients. Methods: In this retrospective study, 192 children undergoing renal ultrasound and DMSA scintigraphy were included. Renal morphometric and volumetric parameters, along with shear wave elastography, were analyzed. Supervised machine learning models were trained to predict renal scarring. A validated data augmentation framework addressed class imbalance and limited sample size. Model performance was assessed using standard classification metrics. Results: Kidney volume indexed to body surface area and the asymmetry index were strongly associated with renal scarring. Elastography alone had limited discriminatory power in conventional analyses but improved predictive performance when incorporated into machine learning models after data augmentation. Ensemble-based models achieved the highest accuracy and area under the receiver operating characteristic curve. Conclusions: Multiparametric ultrasound with machine learning shows potential as a noninvasive tool for predicting renal scarring in children. While not a replacement for DMSA scintigraphy, this approach may aid risk stratification and clinical decision-making, potentially reducing unnecessary radiation exposure.