A novel ct-based radiomics approach for kidney function evaluation in adpkd

Abstract Background and Aims Clinical management of autosomal dominant polycystic kidney disease (ADPKD) might take advantage of the use of new tools to predict risk of progression towards end stage kidney disease (ESKD). The aim of this study is to develop and validate a model based on radiomic features to predict kidney function among patients with ADPKD obtained from CT scans performed for the determination of total kidney volume (TKV). Method We retrospectively selected a cohort of 58 patients with ADPKD who underwent CT scan from February 2020 to March 2021, including 30 patients with eGFR ≥ 60 mL/min/1.73 m2 and 28 with eGFR < 60 mL/min/1.73 m2 at baseline. An expert radiologist generated a region of interest (ROI) segmentation for cystic kidney compounds, obtaining 58 ROIs from which we extracted 217 radiomic features using a dedicated software. We built three different logistic regression models to predict kidney function based on different predictors: height-adjusted TKV (ht-TKV), a selected radiomic feature (F_cm_merged.clust.tend), and both. Area under the curve (AUC) of the receiver operating characteristic (ROC) and accuracy were employed to evaluate models’ performance in discriminating between the two eGFR groups. Internal 3-fold cross-validation (CV) was performed. Results The ht-TKV, radiomic and combined models presented respectively an AUC (95% confidence interval) of 0.79 (0.67, 0.91), 0.83 (0.72, 0.93), 0.84 (0.74, 0.94), confirmed by the CV. Mean (standard deviation) values of the accuracy over CV iterations were 0.67 (0.10), 0.77 (0.09), 0.77 (0.09) for the three models. A model combining ht-TKV with a radiomic feature based on CT images from polycystic kidneys resulted effective in the prediction of baseline kidney function in our cohort. Furthermore, a logistic regression model based on a different radiomic feature (F_cm.2.5Dmerged.info.corr.2) selected among a subcohort of 29 ADPKD patients with a clinical follow-up, predicted rapid progression with a AUC of 0.81, a sensitivity of 100% and a specificity of 53%. Conclusion This is among the first studies which aimed to investigate, in a clinical setting, radiomics potential ability in discriminating eGFR at baseline and to explore as well whether a reliable radiomic feature could be taken into account in predicting faster rapid kidney function impairment over time. Further studies should implement a model extension to predict kidney function slope in order to confirm the role of radiomics in ADPKD management.