A pilot study of magnetic resonance fingerprinting and radiomics analysis in autosomal dominant polycystic kidney disease.

Abstract

This study investigated radiomic features extracted from magnetic resonance fingerprinting (MRF)-acquired T1 and T2 quantitative maps in children and young adults with autosomal dominant polycystic kidney disease (ADPKD) with normal kidney function and compared these features to those extracted from MRF images of healthy volunteers. Mean T1 and T2 relaxation times, along with additional radiomic features, were derived from MRF T1 and T2 maps in 11 healthy volunteers and 18 patients with ADPKD. Two radiomic feature schemes were investigated to differentiate between patients with ADPKD and a healthy cohort using features from either the non-cystic kidney parenchyma or the entire kidney: mean T1 and mean T2 relaxation times alone and feature selection to identify radiomic features from the MRF T1 and T2 maps. Clinical models based on age, height-corrected total kidney volume (htTKV), and body surface area corrected TKV (BSA-TKV) were also investigated. Area under the receiver operating characteristic curve (AUC) was used to evaluate classification performance. MRF-derived mean T1 and T2 relaxation times in the non-cystic kidney parenchyma and the entire kidney was significantly greater for patients with ADPKD than for healthy volunteers. Combining radiomic features allowed for promising differentiation of patients with ADPKD and healthy volunteers using the non-cystic kidney parenchyma (AUC values: 0.79-0.99) and the entire kidney (AUC values: 0.92-0.96). The clinical biomarkers BSA-TKV and htTKV achieved AUC values of 0.86 and 0.72, respectively. There was a strong ability to discriminate between patients with ADPKD and healthy volunteers using features from the non-cystic kidney parenchyma generated by MRF-based mean T2 and a combination of radiomic features extracted from either the MRF T1 or T2 map. Our study suggests that early alterations in non-cystic kidney parenchyma can detect the presence of ADPKD using radiomic features from MRF T1 and T2 maps.