Intra- and interobserver variability in renal multiparametric magnetic resonance imaging: T1 mapping, diffusion-weighted imaging, arterial spin labeling, blood oxygen level-dependent and renal artery flow measurements utilizing manual segmentation

Background: Multiparametric MRI (mpMRI) offers non-invasive characterization of renal microstructure and function, but its clinical translation is hindered by the lack of standardized post-processing. Manual segmentation is widely used but inherently observer-dependent. This study aimed to quantify the intra- and interobserver variability in manual segmentation of a comprehensive renal mpMRI protocol in a diverse cohort of patients with biopsy-confirmed chronic kidney disease (CKD).

Methods: Seventy-two patients scheduled for diagnostic kidney biopsy were prospectively enrolled. Participants underwent mpMRI on a 1.5-T system prior to biopsy, including kidney volume assessment, T1 mapping, T2* [blood oxygen level-dependent (BOLD)], diffusion-weighted imaging [apparent diffusion coefficient (ADC)], arterial spin labeling (ASL), phase-contrast renal artery flow, and T1-rho mapping. Manual segmentation of the renal cortex and medulla was performed on three representative coronal slices. Two trained independent observers (one nephrologist, one radiologist) analyzed the cohort to assess interobserver variability. One observer repeated the full analysis after a six-month washout period to assess intraobserver variability. Agreement was evaluated using intraclass correlation coefficients (ICCs) and Bland-Altman analysis.

Results: Complete datasets were available for 64 patients (mean estimated glomerular filtration rate 44 mL/min/1.73 m²; range, 3–132 mL/min/1.73 m²). Intra- and interobserver ICCs indicated good-to-excellent agreement for T1, T2*, and ADC in both cortical and medullary compartments (ICC range, 0.84–0.99). ASL showed excellent intraobserver reliability (ICC =0.94), but interobserver agreement was only good (ICC =0.78). Renal artery flow measurements also demonstrated good intra- and interobserver agreement (ICC =0.82 for both). T1-rho showed moderate-to-good intraobserver agreement (ICC range, 0.73–0.81) but was limited by image quality.

Conclusions: Manual segmentation of renal cortex and medulla yields highly reproducible measurements for T1, T2*, ADC, renal artery flow and ASL in a mixed CKD population, supporting their validity for clinical research. However, the proportional bias in the Bland-Altmann plots for ASL and renal artery flow highlights the necessity for rigorous standardization of vessel selection and potential automation to ensure reliability across different readers.