利尿超声评估儿童肾盂输尿管连接部梗阻手术必要性

RATIONALE AND OBJECTIVES: This study aimed to develop an interpretable machine learning (ML) model using diuretic ultrasonography to predict the necessity for surgical intervention in children aged 1 month to 18 years with ureteropelvic junction obstruction (UPJO), thereby providing a non-invasive, radiation-free approach to optimize clinical decision-making.

MATERIALS AND METHODS: This study included 41 pediatric patients with UPJO, involving 48 renal units. Participants underwent standardized diuretic ultrasonography, during which the anteroposterior diameter (APD) of the renal pelvis was measured at multiple time points following furosemide injection. Renal excretion curves were derived from sequential APD measurements, and characteristic dynamic parameters were extracted. Key predictors were identified using least absolute shrinkage and selection operator (LASSO) regression and utilized to train six ML models. Model performance was assessed using nested cross-validation, and interpretability was enhanced through SHapley Additive exPlanations (SHAP) analysis.

RESULTS: Excretion curves were fitted based on patient data. LASSO regression identified key predictive features, including the area under the diuretic-APD excretion curve (APD-AUC), descent percentage at 45 min, and Timing Grading of Pelviectasis Regression (TGPR). The support vector machine (SVM) model demonstrated superior performance, achieving an AUC of 0.984. SHAP analysis confirmed that a larger curve area, lower descent rate, and higher TGPR value were strongly associated with the need for surgical intervention. The finalized model was deployed as an interactive web application for clinical implementation.

CONCLUSION: This study developed an ML model using diuretic ultrasound-derived excretion curves to accurately identify optimal surgical timing in pediatric UPJO, offering a practical tool for personalized hydronephrosis management.