mydata.zip

Evaluation of split renal function is crucial for informing clinical decision-making processes. Currently, there are few methods for quantitatively assessing split renal function in clinical practice. The most commonly used technique, renal dynamic imaging, is limited in its widespread application due to radiation exposure and complex procedures. To address these limitations, we developed a radiomics-based evaluation approach. The present investigation encompassed 543 eligible kidneys, which were allocated into training and internal validation sets in a ratio of 7:3. Split renal function glomerular filtration rate from renal dynamic imaging was utilized as the reference standard and was stratified into three categories according to chronic kidney disease staging: <30ml/min/1.73m², 30-45ml/min/1.73m², and >45ml/min/1.73m². In the training set, 1316 radiomics features were extracted from the computed tomography scans of each kidney. Based on the feature importance ranking from the tree-based model, 19 features were selected to construct the random forest model. The model exhibited robust discriminative capabilities in the validation set, with overall discriminative metrics (Macro and Micro AUC [95% CI]: 0.819[0.768-0.864] and 0.832[0.782-0.875]) indicating favorable outcomes. The AUC for the respective categories were 0.914[0.858-0.957], 0.726[0.649-0.800], and 0.818[0.751-0.883]. Calibration curves confirmed the model’s accuracy for each category, while decision curve analysis highlighted its clinical utility. Thus, the radiomics model offers a streamlined alternative for assessing split renal function, especially for patients unsuitable for renal dynamic imaging.