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

Area of Science:

• Radiology
• Medical Imaging
• Urology

Background:

• Accurate differentiation of kidney stone composition is crucial for effective treatment planning.
• Uric acid stones require different management strategies compared to other stone types.

Purpose of the Study:

• To develop and validate a prediction model using non-contrast computed tomography (NCCT) radiomic features to distinguish uric acid stones from non-uric acid stones.
• To assess the model's performance and clinical utility in pre-treatment stone characterization.

Main Methods:

• Retrospective analysis of 195 patients with confirmed renal stone composition.
• Extraction and selection of radiomic features from NCCT images.
• Development of clinical, radiomics, and combined prediction models using machine learning algorithms (Logistic Regression, SVM, MLP, ExtraTrees, LightGBM).

Main Results:

• A combined model integrating radiomic and clinical features achieved a high area under the receiver operating characteristic curve (AUC) of 0.886 in the training set and 0.805 in the test set.
• The combined model demonstrated significantly superior performance compared to the clinical and radiomics-only models.
• Shapley additive explanations (SHAP) analysis highlighted the importance of texture features in predicting uric acid stones.

Conclusions:

• The developed combined model based on NCCT radiomics shows strong performance in differentiating uric acid stones.
• This model can serve as a valuable tool for guiding pre-treatment decisions and personalizing patient management.