Prediction of kidney function decline in patients with autosomal dominant polycystic kidney disease using radiomic features extracted from the non-cystic kidney parenchyma

Autosomal dominant polycystic kidney disease (ADPKD) is the most common genetic kidney disease leading to the gradual enlargement of kidneys over time due to cyst growth, ultimately resulting in loss of kidney function; 50% of ADPKD patients enter end-stage kidney disease (ESKD) by 56 years of age. Radiologic imaging is crucial in monitoring ADPKD patients to measure total kidney volume (TKV) for risk to ESKD. Radiomic models have shown power to predict loss of kidney function or chronic kidney disease (CKD) stage. Potential additive value is obtained from utilizing non-cystic kidney parenchyma texture beyond typical clinical measures such as TKV or estimated glomerular kidney filtration rate (eGFR). Longitudinal magnetic resonance (MR) imaging has been used to clinically monitor patients; therefore, a wealth of imaging timepoints can be incorporated in predictive models developed from the non-cystic kidney parenchyma. We studied a cohort of 94 patients matched for age, sex, and eGFR along with their respective T2-weighted fat saturated MR imaging coronal images to predict progression to CKD stage 3A or greater after 60 months using radiomic features extracted from baseline and 24- and 48-months follow-up MR images. Prediction models yielded area under the receiver operating characteristic curve (AUC) values that ranged from 0.649-0.721. Not one model was statistically significantly different from the baseline texture feature model alone. Radiomic features from the non-cystic kidney parenchyma extracted from MR images reveal prognostic power in predicting kidney function decline and warrant further investigation.