An interpretable machine learning model for preoperative prediction of renal mass malignancy.

objectiveTo develop and validate a predictive model for distinguishing benign and malignant renal masses using machine learning (ML) algorithms.

methodsWe analyzed data from 1084 patients diagnosed with renal masses between June 2020 and November 2023. Patients were randomly divided into training and validation cohorts at an 8:2 ratio. Seven ML algorithms were employed to construct prediction models for malignant tumors. The area under the receiver operating characteristic curve (AUC) was used as the primary evaluation metric to identify the best model. Comprehensive model evaluation included AUC, accuracy, sensitivity, specificity, and other performance indicators, with validation on an independent cohort. The SHapley Additive exPlanation (SHAP) algorithm was applied to calculate and visualize feature importance.

resultsEight variables were included in the final model: gender, cystic/solid nature of the mass, maximum mass diameter, uric acid levels, endogenous creatinine clearance, albumin, albumin-to-globulin ratio, and direct/total bilirubin. Among the seven ML models, the Gradient Boosting Decision Tree (GBDT) model performed best, achieving an AUC of 0.88 (95% CI 0.82-0.95), accuracy of 0.80, sensitivity of 0.82, and specificity of 0.79 in the validation cohort. The SHAP algorithm revealed that the cystic/solid nature of the mass was the most significant factor influencing the model's predictions. A publicly accessible web application was developed to facilitate patients.

conclusionsThe GBDT model demonstrates high overall effectiveness and offers substantial support for the clinical differentiation of benign and malignant renal masses.