Abstract
ABSTRACT
Determine whether non‐contrast MRI metrics of kidney structure and hemodynamics predict glomerular filtration rate (GFR). Develop and prospectively validate an image‐based estimated GFR (eGFR) approach that is free of radiation and serum biomarkers. Eighty‐nine healthy adults without known kidney disease underwent a 3 T MRI exam, including T1‐weighted anatomical imaging, phase contrast MRI for renal blood flow (RBF), and T2* mapping. Seventy‐four participants comprised a model development cohort and 20 served as a prospective validation cohort. Serum creatinine‐based eGFR was used as the reference standard. Bivariate and multivariate linear regression identified imaging predictors of eGFR and was used prospectively to generate image‐based eGFR models. Models were evaluated with bootstrapping and prospectively validated using bias, root means squared error, and p30/p20/p10 accuracy metrics. Age‐related trajectories of imaging parameters were also assessed. On bivariate analysis, cortical thickness (
r
2
= 0.38), RBF (
r
2
= 0.29), total kidney volume, cortical R2*, and medullary R2* were significantly associated with eGFR (all
p
r
2
= 0.47,
p
2
) and acceptable accuracy (p30 ≈87%–92%) in both bootstrapped testing and prospective validation. In our healthy cohort, age was the strongest determinant of eGFR, RBF, and cortical thickness, with similar age‐related declines in eGFR (−8.0%/decade) and RBF (−7.8%/decade). Non‐contrast MRI measures of renal cortical thickness and blood flow are strong, complementary predictors of kidney function. An image‐based eGFR model achieved bias and accuracy comparable to serum creatinine–based equations in healthy adults. This proof‐of‐concept supports the feasibility of a safe, contrast‐free, and physiology‐informed approach to estimating GFR and motivates further validation against measured GFR and in broader clinical populations.