Radiomics application using non-contrast computed tomography for predicting uric acid kidney stones.

objectiveThis study aims to develop a prediction model based on non-contrast computed tomography (NCCT) images to differentiate uric acid stones from non-uric acid stones before treatment.

methodsThis study retrospectively enrolled 195 patients from Quzhou People's Hospital between 2022 and 2024 who underwent dual-energy CT scans with confirmed renal stone composition. The patients were randomly divided into a training set (156 cases) and a test set (39 cases) in an 8:2 ratio. Regions of interest (ROIs) were manually delineated slice-by-slice on NCCT images to extract radiomic features. Feature dimensionality reduction and selection were performed using intraclass correlation coefficient (ICC), Spearman rank correlation coefficients, and least absolute shrinkage and selection operator (LASSO) regression. Radiomics and clinical models were developed using logistic regression (LR), support vector machine (SVM), multilayer perceptron (MLP), ExtraTrees, and LightGBM algorithms. Finally, a combined model was constructed by integrating the selected radiomic features with clinically significant risk factors. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), while clinical utility was assessed through decision curve analysis (DCA). Model interpretability was examined using Shapley additive explanations (SHAP).

resultsA total of 1834 radiomic features were extracted from each ROI. After feature dimensionality reduction and selection, 6 radiomic features remained for characterizing stone composition. The clinical, radiomics, and combined models all demonstrated favorable discriminatory power for uric acid stones. The AUC values of the three models in the training set were 0.744 (95% CI: 0.666-0.822), 0.831 (95% CI: 0.766-0.897), and 0.886 (95% CI: 0.834-0.938), respectively, and 0.778 (95% CI: 0.631-0.925), 0.634 (95% CI: 0.454-0.777), and 0.805 (95% CI: 0.666-0.944) in the test set. DeLong's test indicated that in the training set, the performance of the combined model was significantly superior to both the clinical model and the radiomics model (0.886 vs. 0.744, p < 0.001; 0.886 vs. 0.831, p = 0.015). Decision curve analysis also demonstrated its potential clinical utility. SHAP analysis revealed that texture features were important factors in predicting uric acid stones.

conclusionThe combined model based on NCCT performs well in distinguishing uric acid stones and can provide effective references for treatment decisions. CLINICAL TRIAL NUMBER: Not applicable.