MRI-Based Radiomics to Predict Renal Function Response to Renal Artery Stenting for Atherosclerotic Renal Artery Stenosis.

Abstract

Purpose: To investigate magnetic resonance imaging (MRI)-based radiomics for predicting renal function response for patients treated for atherosclerotic renal artery stenosis (ARAS) by endoluminal means.

Material and methods: A cohort of 146 ARAS patients who underwent stenting was analyzed, with retrospective training and prospective validation groups delineated based on the treatment timing. Patients were categorized into benefit and no-benefit groups based on postoperative renal function during follow-up. Optimal radiomics labels were selected from regions of interest (ROIs) including the stenotic side and both kidneys. The nomogram combined optimal radiomics signatures with independent clinical factors using multivariable logistic regression. Shapley Additive exPlanations (SHAP), decision curve analysis (DCA), the net reclassification index (NRI), and the total integrated discrimination index (IDI) were conducted to determine the clinical usefulness of the nomogram.

Results: Split renal function of the stenotic side and diabetes emerged as independent clinical predictors. A nomogram, incorporating these clinical factors and radiomics features from the stenotic side and both kidneys, achieved area under the curve (AUCs) of 0.927 (0.861-0.979) and 0.904 (0.819-0.972) in the training and test groups, respectively, for predicting benefits. The clinical-radiomics model significantly improved diagnostic performance (p = 0.001 and p = 0.011 for the training and test groups, respectively). DCA, NRI, and IDI analyses suggested the nomogram's superiority. SHAP analysis highlighted the radiomics feature from stenotic side kidney as the most critical predictive feature.

Conclusions: Both MRI radiomics and clinical factors may be valuable in pre-treatment counseling of ARAS patients who may benefit from endovascular treatment.