Virtual MR elastography with diffusion-weighted imaging to evaluate chronic kidney disease fibrosis and dysfunction

Abstract

Background

Renal stiffness declines with chronic kidney disease (CKD) progression. Magnetic resonance elastography (MRE) enables quantitative assessment of tissue stiffness but requires external hardware, limiting routine clinical use.

Purpose

To develop a diffusion-derived virtual shear modulus (µ_diff) for vibration-free renal virtual MR elastography (vMRE) and evaluate its ability to assess dysfunction and fibrosis across CKD stages.

Materials and Methods

This prospective single-center study included 53 healthy volunteers and 114 CKD patients, stratified by estimated glomerular filtration rate (eGFR) into early-stage CKD (CKD_1-2, eGFR 60-120 mL/min/1.73 m²) and moderate-to-advanced CKD (CKD_3-5, eGFR < 60 mL/min/1.73 m²). Renal biopsy results performed for clinical care were collected. Magnetic resonance elastography and multi-b-value diffusion-weighted imaging (DWI) (b = 0-2000 s/mm²) were performed on a 3T MRI (MAGNETOM Prisma, Siemens). Cortical shifted apparent diffusion coefficient (csADC) across b-value combinations correlated with cortical MR elastographic shear modulus (cµ_MRE) to identify the optimal combination and derive µ_diff. Age-adjusted analysis of covariance (ANCOVA) compared MRI parameters. Partial correlations with eGFR and Spearman correlations with interstitial fibrosis (IF) were performed. Diagnostic performance was evaluated by receiver operating characteristic (ROC) analysis.

Results

A significant correlation (R² = 0.94, P < .001) was observed between cµ_MRE and csADC derived from DWI at b-values of 200 (S₂₀₀) and 800 (S₈₀₀) s/mm² (csADC_200-800), yielding ${\mathrm{\mu }}_{\mathrm{diff}} \left(\mathrm{kPa}\right)=4.91 \ln (S_{200}/S_{800})+0.92.$ Cortical µ_diff (cμ_diff) correlated with eGFR (r = 0.602, P < .001) and IF (r = −0.809, P < .001). In biopsied patients (n = 51), cμ_diff showed comparable diagnostic performance to cµ_MRE in differentiating mild from moderate-to-severe IF (area under the curve [AUC], 0.910 [95% CI, 0.822-0.998] vs 0.901 [95% CI, 0.813-0.989]; P > .05).

Conclusion

Diffusion-weighted imaging-derived µ_diff enables vibration-free renal vMRE, offering a stiffness-related surrogate that correlates with renal dysfunction and fibrosis. Diagnostic performance is comparable to MRE, including in early-stage CKD. If validated externally, this may provide a scalable alternative for CKD evaluation.

Abbreviations 

BMI = body mass index, CKD = chronic kidney disease, csADC = cortical shifted apparent diffusion coefficient, DWI = diffusion-weighted imaging, eGFR = estimated glomerular filtration rate, FOV = field of view, MRE = magnetic resonance elastography, ROC = receiver operating characteristic, ROI = region of interest, SWS = shear wave speed

Summary 

Renal virtual magnetic resonance elastography using diffusion-weighted imaging enables vibration-free evaluation of stiffness-related tissue alterations with outputs that are correlated with renal function and fibrosis, including early-stage chronic kidney disease.

Key Results 

• Shifted apparent diffusion coefficient from diffusion-weighted imaging at b-values of 200 − 800 s/mm² showed the strongest correlation with MR elastographic shear modulus (R² = 0.94, P < .001), yielding a stiffness-related surrogate equation.

• Virtual MR elastography correlated with renal function and fibrosis (r_max = 0.602, −0.807; P < .001) and discriminated stiffness-related alterations in early chronic kidney disease.

• Pathologically validated virtual MR elastography showed diagnostic performance comparable to MR elastography for renal fibrosis.

Introduction

Chronic kidney disease (CKD) is a global disorder causing progressive renal decline to end-stage kidney disease (ESKD).¹ Key histopathologic features include glomerulosclerosis (GS), interstitial fibrosis (IF), and tubular atrophy (TA), with IF as the main driver of CKD progression.² Accurate assessment of renal structural injury is essential for early detection and prognosis. Clinically, estimated glomerular filtration rate (eGFR), derived from serum creatinine (Scr), indirectly reflects overall renal function but is not sensitive to early localized pathology. It shows marked alterations only in advanced disease and is affected by hemodynamics.³ Renal biopsy remains the reference standard but is invasive and unsuitable for longitudinal monitoring. Therefore, noninvasive, repeatable, cost-effective, and spatially informative imaging techniques are crucially needed to detect early structural changes and monitor CKD progression.

Magnetic resonance elastography (MRE) is an imaging biomarker that quantifies tissue stiffness using shear wave speed (SWS) or MR elastographic shear modulus (μ_MRE). Magnetic resonance elastography has demonstrated 90%-99% sensitivity and specificity in hepatic fibrosis assessment⁴ and has been increasingly applied to renal evaluation. Compared with eGFR and biopsy, MRE provides a noninvasive and spatially resolved assessment of renal stiffness. Several studies confirm the feasibility of MRE for renal disease and suggest higher sensitivity in the cortex than the medulla for detecting early CKD impairment.^5–8 However, MRE is limited by the need for specialized hardware and intolerance to mechanical vibration in some patients (eg, post-biopsy or polycystic kidneys). These factors hinder its routine and longitudinal use.

To overcome these challenges, Le Bihan et al.⁹ proposed a diffusion-weighted imaging (DWI)-based technique, virtual MR elastography (vMRE). This method uses 2 key b-values (200, 1500 s/mm²) to convert the shifted apparent diffusion coefficient (sADC) into diffusion-derived virtual shear modulus (μ_diff), enabling diffusion-based estimation of stiffness-related tissue properties without mechanical vibration. Virtual MR elastography is feasible in various organs (eg, liver, placenta, breast, lung),^4,9–11 showing good correlation with histopathology. Diffusion-weighted imaging is sensitive to tissue microstructure, especially at higher b-values where restricted diffusion captures non-Gaussian information.¹² Compared with MRE, vMRE offers greater accessibility and efficiency, providing diffusion-derived metrics associated with tissue stiffness for quantitative assessment. Using routine DWI sequences without extra equipment, vMRE is potentially scalable, particularly in resource-limited settings.¹²

Unlike chronic liver disease, where stiffness increases with fibrosis progression, renal stiffness decreases with CKD progression,⁵ indicating organ-specific biomechanical behavior. Therefore, liver-derived diffusion models may not directly apply to the kidney. A previous study showed that sADC derived at high b-values (800 s/mm²) has clinical value for noninvasive CKD evaluation.¹³ However, the optimal b-value combination for renal vMRE remains undefined. Because b-value selection substantially affects diffusion sensitivity, higher b-values (≥800 s/mm²) enhance diffusion weighting and reduce perfusion effects.¹⁴

Our primary hypothesis was that sADC correlates with MRE-derived shear modulus (µ_MRE). Based on this relationship, we aimed to derive a diffusion-derived virtual shear modulus (μ_diff) reflecting stiffness-related properties for renal vMRE. Our secondary hypothesis was that vMRE would achieve diagnostic performance comparable to conventional MRE in assessing CKD-related structural alterations and disease staging, providing a simple, reproducible, and hardware-free approach for renal assessment.

Materials and methods

Study participants

This prospective single-center study complied with the Declaration of Helsinki and was approved by the Ethics Committee of Shenzhen Traditional Chinese Medicine Hospital (K2023-091). Written informed consent was obtained from all participants. Sample size calculation using G*Power (version 3.1.9.7) is detailed in the Supplementary Materials. From March 2023 to July 2025, 131 CKD patients were prospectively recruited from the nephrology outpatient clinic and inpatient department of the study center. During the same period, 61 healthy volunteers (HVs) were recruited from hospital staff and their family members based on prior health examination records and self-reported medical history. Chronic kidney disease was defined as abnormalities of kidney structure or function persisting for ≥3 months according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria.¹⁵ Exclusion criteria included: (1) kidney transplantation; (2) renal malignancy; (3) large (>2 cm) or multiple benign renal lesions (eg, angiomyolipoma or polycystic kidney disease); (4) renal calculi with hydronephrosis; and (5) poor image quality with substantial artifacts. Chronic kidney disease patients were stratified into early-stage CKD (CKD_1-2) and moderate-to-advanced CKD (CKD_3-5) based on the clinically relevant eGFR threshold of 60 mL/min/1.73 m².¹⁵ The inclusion criteria for HVs were: (1) no history of renal disease, diabetes, hypertension, or vascular disease; and (2) no nephrotoxic drug use in the past 3 months.

MRI examination

All participants fasted and abstained from water for 4 hours before MRI. Imaging was performed on a 3T scanner (MAGNETOM Prisma, Siemens Healthineers, Forchheim, Germany) using an 18-channel phased-array body coil in the supine position. Participants were positioned supine and instructed to maintain steady respiration. The protocol included coronal MRE and multi-b-value DWI sequences. Magnetic resonance elastography parameters: free breathing; repetition time (TR)/echo time (TE), 2450/56 ms; voxel size, 2.5 × 2.5 × 2.5 mm³; slice thickness, 2.5 mm; field of view (FOV), 250 × 250 mm²; driving frequencies, 40, 50, 60, and 70 Hz; acquisition time, 4 min 24 s. Diffusion-weighted imaging parameters: TR/TE, 2700/48 ms; voxel size, 2.5 × 2.5 × 4.0 mm³; slice thickness, 4 mm; FOV, 250 × 250 mm²; b-values, 0, 50, 100, 200, 400, 600, 800, 1000, 1200, 1500, and 2000 s/mm²; acquisition time, 4 min 41 s.

MR imaging data acquisition and processing

MRE data post-processing and quantification of μ_MRE

Magnetic resonance elastography data were post-processed using the wave-number-based multi-frequency dual elasto-visco (k-MDEV) reconstruction algorithm (version 1.0171; available at www.bioqic-apps.charite.de), as previously described,^16,17 to generate elastograms (cMap) and magnitude (Mag) images. The Mag image is a T2-weighted magnitude image showing renal anatomy, while the cMap provides SWS (m/s), which reflects tissue stiffness. The MR elastographic shear modulus (μ_MRE) was calculated as:^18,19

$${\mathrm{\mu }}_{\mathrm{MRE}}\left(\mathrm{kPa}\right)=\rho {\mathrm{SWS}}^2,\mathrm{ }\rho =1000\mathrm{ }\mathrm{kg}/{\mathrm{m}}^3\mathrm{ }$$ (1)

Diffusion MR imaging-sADC generation and b-value combination selection

The sADC was calculated using 2 b-values (low b-value, LKb, and high b-value, HKb) to capture both Gaussian and non-Gaussian diffusion:⁹

$$\mathrm{sADC}\mathrm{ }(\mathrm{m}{\mathrm{m}}^2/s)=\ln \left(S_{\mathrm{LKb}}/S_{\mathrm{HKb}}\right)/(\mathrm{HKb}-\mathrm{LKb})\mathrm{ }$$ (2)

where S_LKb and S_HKb denote signal intensities at LKb and HKb, respectively. As the renal cortex is more sensitive to CKD progression than the medulla,⁵ it was chosen as the primary region of analysis. In advanced CKD, cortical thinning and loss of the cortico-medullary boundary hinder reliable measurements; therefore, patients with CKD stages 1-3 (n = 75) were pooled to improve statistical robustness. Based on these considerations, 7 b-value combinations (200-800, 200-1000, 200-1200, 200-1500, 200-2000, 400-1200, 400-1500 s/mm²) were evaluated for sADC-μ_MRE correlation. The rationale for b-value selection, including exclusion of very low b-values to minimize perfusion-related effects, is provided in the Supplementary Materials. Shifted apparent diffusion coefficient images at different b-value combinations were generated using MATLAB (Mathworks).

Diffusion-based vMRE-estimating virtual shear modulus

After identifying the b-value combination with the strongest sADC-μ_MRE correlation, the relationship was further explored using scatter plots. Inverting this relationship, μ_diff, could be estimated from the sADC obtained with those b-values:⁹

$${\mathrm{\mu }}_{\mathrm{diff}}\left(\mathrm{kPa}\right)=g\left[\ln \left(S_{\mathrm{LKb}}/S_{\mathrm{HKb}\mathrm{ }}\right)\right]\mathrm{ }$$ (3)

where g denotes an empirically determined linear regression transform used to establish a kidney-specific μ_diff model converting DWI attenuation into a diffusion-derived virtual shear modulus. Subsequently, μ_diff maps were generated using custom-written software in MATLAB, with negative voxel values excluded from analysis.

Region of interest acquisition

Post-processing generated sADC, μ_diff, and cMap parameter maps (Figure S1). Two experienced abdominal radiologists (reader 1 with 10 years and reader 2 with 5 years of MRI experience), blinded to clinical and histopathologic data, independently delineated regions of interest (ROIs) according to established methods.¹⁷ Regions of interest were placed on DWI (b = 0 s/mm²) and magnitude images using 3 consecutive slices centered at the renal hilum. On each slice, one cortical ROI was delineated in each kidney along the outer cortex (∼70-110 voxels per ROI), avoiding vessels and artifacts.²⁰ In total, 6 cortical ROIs (3 per kidney) were obtained for analysis. Additional medullary ROIs (∼16 voxels each) were placed in the upper, middle, and lower regions, avoiding the corticomedullary junction and renal sinus vasculature (Figure S1).^13,21 All ROIs were manually delineated using ITK-SNAP²² and co-registered with sADC, μ_diff, and cMap images for quantitative analysis. Bilateral values were averaged to reduce inter-kidney variability. Inter- and intra-reader reproducibility were assessed, with reader 1 repeating measurements after 1 month.

Renal histopathology

Fifty-one CKD patients who underwent ultrasound-guided kidney biopsy as part of routine clinical care based on clinical indications were included in the histopathologic analysis, with MRI performed within 7 days. The biopsy site was the lower pole of the left or right kidney in our study. The specimens were processed with PASM and Masson staining.¹³ Chronic lesions were graded according to a published scoring system,²³ including global and segmental GS (0-3 points), TA (0-3 points), IF (0-3 points), and arteriosclerosis (CV, 0-1 point). The total score was used for grading: minimal (0-1), mild (2-4), moderate (5-7), and severe (8-10), and IF additionally stratified by cortical involvement as mild (≤25%), moderate (26%-50%), or severe (>50%). Biopsies were reviewed by 2 experienced renal pathologists blinded to imaging data to ensure diagnostic consistency.

Statistical analysis

Statistical analyses were performed using SPSS (version 26.0, IBM Corp., Armonk, NY, United States), GraphPad Prism (version 9.0.0, San Diego, CA, United States), MedCalc (version 20.022), and Chiplot (https://www.chiplot.online/). Intraclass correlation coefficients (ICCs) assessed inter- and intra-reader agreement. The first reader’s results from the first assessment were used for analysis. Normally distributed data were expressed as mean ± standard deviation and non-normally distributed data as median (interquartile range [IQR]). Sex ratio was compared using the chi-square test. The Mann-Whitney U test was applied to compare eGFR and Scr between CKD_1-2 and CKD_3-5 groups. Pearson correlation identified the optimal b-value combination for µ_diff regression. Analysis of covariance (ANCOVA), adjusted for age with Bonferroni correction, compared vMRE and MRE parameters among HV, CKD_1-2, and CKD_3-5 groups. Spearman correlation analyzed relationships of vMRE and MRE parameters with pathological indices. Partial correlation, adjusted for age, body mass index (BMI), and sex, evaluated relationships with renal function indicators. Receiver operating characteristic (ROC) analysis was performed to assess diagnostic performance for differentiating HV from CKD_1-2, CKD_1-2 from CKD_3-5, and mild from moderate-to-severe IF. Pairwise comparisons of areas under the curve (AUCs) were performed using Bonferroni-adjusted DeLong tests. Cutoff values were derived from ROC analysis using the Youden index, defined as sensitivity + specificity − 1, to determine the threshold with the optimal balance between sensitivity and specificity. A 2-sided P < .05 was considered statistically significant.

Code sharing

The MATLAB code used for generating the sADC and μ_diff maps is publicly available at https://anonymous.4open.science/r/virtalMRE-3210/README.md

Results

Clinical characteristics

A total of 192 participants were initially enrolled (Figure 1). Twenty-five participants were excluded because of poor image quality (n = 10), renal calculi with hydronephrosis (n = 1), renal cysts > 2 cm (n = 6), apparent renal atrophy (n = 2), incomplete clinical data (n = 4), and patient refusal (n = 2). Ultimately, 53 HVs and 114 CKD patients were included, subdivided into CKD_1-2 (n = 42) and CKD_3-5 (n = 72) groups (Table 1). Age differed significantly among the HV, CKD_1-2 and CKD_3-5 groups (P = .006), whereas sex distribution (male/female: HV, 25/28; CKD_1-2, 25/17; CKD_3-5, 46/26; P > .05) and BMI showed no significant differences (P > .05). Among CKD patients, 51 underwent renal biopsy and were stratified by the severity of IF into mild (n = 23) and moderate-to-severe (n = 28) groups. All parameters exhibited high consistency (Table S1).

Figure 1.

Flowchart of study participants. Abbreviations: CKD, chronic kidney disease; HV, healthy volunteers.

Table 1.

Demographics and clinical data.

	HV	CKD1 - 2	CKD3 - 5	P
Numbers (n)	53	42	72	NA
Ages	49 (28.5, 55)	37 (32, 47)	46.69 ± 12.02	.006
Sex, male (%)	25 (47.17%)	25 (59.52%)	46 (63.89%)	.166
BMI (kg/m2)	23.51 ± 2.57	23.76 ± 3.79	23.73 ± 3.54	.917
eGFR (mL/min/1.73 m2)	NA	74 (67, 84)	32.07 ± 7.23	<.001
Scr (μmol/L)	NA	97.79 ± 21.46	192 (138, 315)	<.001

Values are presented as mean ± SD for normally distributed variables and as median (IQR) for non-normally distributed variables.

Abbreviations: BMI, body mass index; CKD, chronic kidney disease; CKD_1-2, CKD stage 1-2; CKD_3-5, CKD stage 3-5; eGFR, estimated glomerular filtration rate; HV, healthy volunteers; NA, not applicable; Scr, serum creatinine.

Pearson correlation between cortical shifted apparent diffusion coefficient and cμ_MRE for optimal b-value selection and μ_diff derivation

The strongest correlation between cortical shifted apparent diffusion coefficient (csADC) and cμ_MRE occurred at the b-value combination of 200-800 s/mm² (R² = 0.94, P < .001) (Figure 2 and Table S2). Linear regression yielded the following equation:

Figure 2.

Correlation between csADC_200-800 and cμ_MRE. A significant positive correlation was observed between csADC_200-800 and cμ_MRE in patients with CKD stages 1-3 (n = 75) (R² = 0.94, P < .001), indicating a strong linear relationship. Abbreviations: c, renal cortex; CKD, chronic kidney disease; sADC (10⁻³ mm²/s), shifted apparent diffusion coefficient; μ_MRE (kPa), MR elastographic shear modulus.

$$\mathrm{sADC}\left({10}^{-3} \mathrm{m}{\mathrm{m}}^2/\mathrm{s}\right)=0.32\mathrm{ }{\mathrm{\mu }}_{\mathrm{MRE}}-0.20\mathrm{ }$$ (4)

Integrating this relationship with Equation (4), an empirical μ_diff model was derived:

$${\mathrm{\mu }}_{\mathrm{diff}}\left(\mathrm{kPa}\right)=\mathrm{\alpha }\ln \left(S_{200}/S_{800}\right)+\mathrm{\beta }\mathrm{ }$$ (5)

The scaling (α) and the shift (β) factors were set to 4.91 ± 0.15 and 0.92 ± 0.13, respectively.

Correlations of vMRE and MRE parameters with renal function and histopathological indices

Among 114 CKD patients (Figure 3), vMRE (sADC and μ_diff) and MRE (µ_MRE) correlated positively with eGFR (r = 0.452-0.602, P < .001) and negatively with Scr (r = −0.388 to −0.521, P < .001), with relatively higher correlations observed for vMRE parameters. In the 51 biopsy-proven CKD patients, the vMRE parameter (csADC) exhibited the strongest negative correlation with chronic lesion score (r_max = −0.850, P < .001). Magnetic resonance elastography (µ_MRE) correlated with histopathologic indices, showing a stronger correlation with GS but weaker correlations with chronic lesion score, IF, TA, and CV compared with vMRE parameters.

Figure 3.

Heatmap showing correlations of sADC_200-800, μ_diff200-800, and μ_MRE with clinical indicators (eGFR [n = 114], Scr [n = 114]) and pathological indicators (chronic lesion score, GS, IF, TA, and CV; all n = 51). Deeper colors indicate stronger correlations, and lighter colors indicate weaker correlations. Red denotes positive correlations, and blue denotes negative correlations. Circle size is proportional to the absolute value of the correlation coefficient; larger circles indicate stronger correlations, whereas smaller circles indicate weaker correlations. Virtual MR elastography parameters (sADC, μ_diff) and the MRE parameter (μ_MRE) were positively correlated with eGFR and negatively correlated with Scr and the pathological indicators listed above. Abbreviations: c, renal cortex; CKD, chronic kidney disease; CV, vascular lesions; eGFR, estimated glomerular filtration rate; GS, glomerulosclerosis; IF, interstitial fibrosis; k, kidney parenchyma; m, renal medulla; MRE, MR elastography; sADC (10⁻³ mm²/s), shifted apparent diffusion coefficient; Scr, serum creatinine; TA, tubular atrophy; vMRE, virtual MR elastography; μ_diff (kPa), diffusion-derived virtual shear modulus; μ_MRE (kPa), MR elastographic shear modulus. *P < .05, **P < .01, ***P < .001.

Comparisons of vMRE and MRE parameters among HV, CKD_1-2, and CKD_3-5 groups

As shown in Table 2, Figures 4 and 5, vMRE (sADC and μ_diff) and MRE (μ_MRE) differed significantly among the HV, CKD_1-2, and CKD_3-5 groups (P < .001). Pairwise comparisons demonstrated significant differences in all parameters, except for medullary μ_MRE (mμ_MRE) between HV and CKD_1-2 groups (P > .05), with a gradual decrease across the HV, CKD_1-2 and CKD_3-5 groups.

Table 2.

Comparison of sADC, μ_diff values at b-value combinations of 200-800 s/mm², and µ_MRE values.

	HV	CKD1 - 2	CKD3 - 5	P
ksADC	1.555 ± 0.062	1.505 ± 0.080a	1.391 ± 0.094a,b	<.001
kμdiff	5.502 ± 0.182	5.357 ± 0.235a	5.029 ± 0.275a,b	<.001
kμMRE	6.121 ± 0.498	5.768 ± 0.455a	5.013 ± 0.773a,b	<.001
csADC	1.572 ± 0.072	1.500 ± 0.084a	1.371 (1.307, 1.456)a,b	<.001
cμdiff	5.555 ± 0.205	5.345 ± 0.250a	4.974 (4.784, 5.209)a,b	<.001
cμMRE	5.598 ± 0.377	5.319 ± 0.259a	4.842 (4.520, 5.221)a,b	<.001
msADC	1.491 ± 0.055	1.451 ± 0.074a	1.368 ± 0.084a,b	<.001
mμdiff	5.312 ± 0.162	5.195 ± 0.218a	4.926 ± 0.238a,b	<.001
mμMRE	5.304 (5.132, 5.380)	5.282 (5.167, 5.334)	5.076 (4.384, 5.238)a,b	<.001

Values are presented as mean ± SD for normally distributed variables and as median (IQR) for non-normally distributed variables. P values indicate the overall difference among healthy volunteers.

^avs HV, P < .05.

^bvs CKD_1-2, P < .05.

Abbreviations: c, renal cortex; CKD, chronic kidney disease; CKD_1-2, CKD_3-5 groups with age as a covariate; CKD_1-2, CKD stage 1_-2; CKD_3-5, CKD stage 3-5; HV, healthy volunteers; k, kidney parenchyma; m, renal medulla; sADC (10⁻³ mm²/s), shifted apparent diffusion coefficient; μ_diff (kPa), diffusion-derived virtual shear modulus; μ_MRE (kPa), MR elastographic shear modulus.

Figure 4.

Comparison of MRI parameters among HV, CKD_1-2, and CKD_3-5 groups. Box plots of sADC_200-800 (A), μ_diff200-800 (B), and μ_MRE (C) showed significant differences across most parameters among the 3 groups. Post-hoc analysis indicated that all group comparisons were significant, except for medullary μ_MRE (mμ_MRE) between HV and CKD_1-2, which showed no significant difference. Abbreviations: CKD, chronic kidney disease; CKD_1-2, stage 1_-2 CKD; CKD_3-5, stage 3-5 CKD; HV, healthy volunteers; sADC (10⁻³ mm²/s), shifted apparent diffusion coefficient; μ_diff (kPa), diffusion-derived virtual shear modulus; μ_MRE (kPa), MR elastographic shear modulus. *P < .05, **P < .01, ***P < .001.

Figure 5.

Comparative visualization maps of DWI, sADC_200-800, and μ_diff200-800 in HV, CKD_1-2, and CKD_3-5 groups. Representative cases included a 54-year-old male HV (A, D, G), a 39-year-old male with CKD stage 2 (B, E, H), and a 38-year-old male with CKD stage 5 (C, F, I). DWI at b = 0 s/mm² illustrates overall renal structure (A-C). The sADC_200-800 maps (D-F) show a progressive decline from HV to CKD_3-5 with deeper red indicating higher diffusion and lighter red indicating more restricted diffusion. μ_diff200-800 maps (G-I) display a similar decreasing trend, where deeper red indicates higher μ_diff values, reflecting greater stiffness-related renal tissue changes, and lighter red indicates lower values. Abbreviations: CKD, chronic kidney disease; CKD_1-2, stage 1_-2 CKD; CKD_3-5, stage 3-5 CKD; DWI, diffusion-weighted imaging; HV, healthy volunteers; sADC (10⁻³ mm²/s), shifted apparent diffusion coefficient; μ_diff (kPa), diffusion-derived virtual shear modulus.

Diagnostic performance of vMRE and MRE for CKD staging and interstitial fibrosis

Virtual MR elastography parameters showed moderate diagnostic performance for differentiating HV from CKD_1-2 (AUCs, 0.680 [95% CI, 0.568-0.793]-0.748 [95% CI, 0.648-0.847]), whereas MRE ranged from 0.547 [95% CI, 0.431-0.664] to 0.745 [95% CI, 0.646-0.844]. Notably, the diagnostic performance of mμ_MRE was not statistically significant (AUC, 0.547 [95% CI, 0.431-0.664], P > .05). DeLong tests revealed no significant difference between vMRE and MRE (P > .05) (Table 3). In 51 biopsy-proven CKD patients (Table 4), cμ_diff achieved a slightly higher AUC than cμ_MRE for distinguishing mild from moderate-to-severe IF, without statistical significance (AUCs, 0.910 [95% CI, 0.822-0.998] vs 0.901 [95% CI, 0.813-0.989]; P > .05).

Table 3.

Diagnostic performance of sADC, µ_diff, and µ_MRE in distinguishing different degrees of renal dysfunction in CKD patients.

	AUC (95% CI)	Sensitivity% (95% CI)	Specificity% (95% CI)	Cutoff value	P
HV (n = 53) vs CKD1-2 (n = 42)
ksADC (10−3 mm2/s)	0.717 (0.608-0.825)	57.14 (0.410-0.723)	86.79 (0.747-0.945)	1.499	.421a
kμdiff (kPa)	0.714 (0.605-0.823)	57.14 (0.410-0.723)	86.79 (0.747-0.945)	5.350	.780b
kμMRE (kPa)	0.696 (0.590-0.802)	76.19 (0.605-0.879)	56.60 (0.423-0.702)	6.073	.753c
csADC (10−3 mm2/s)	0.748 (0.648-0.848)	57.14 (0.410-0.723)	90.57 (0.793-0.969)	1.500	.076a
cμdiff (kPa)	0.740 (0.639-0.841)	54.76 (0.387-0.702)	90.57 (0.793-0.969)	5.338	.902b
cμMRE (kPa)	0.745 (0.646-0.845)	61.90 (0.456-0.764)	81.13 (0.680-0.906)	5.362	.951c
msADC (10−3 mm2/s)	0.682 (0.569-0.795)	66.67 (0.505-0.804)	71.70 (0.577-0.832)	1.467	.192a
mμdiff (kPa)	0.680 (0.567-0.793)	66.67 (0.505-0.804)	71.70 (0.577-0.832)	5.241	.058b
mμMRE (kPa)	0.547 (0.430-0.664)	76.19 (0.605-0.879)	37.74 (0.248-0.521)	5.329	.055c
CKD1-2 (n = 42) vs CKD3-5 (n = 72)
ksADC (10−3 mm2/s)	0.819 (0.741-0.898)	68.06 (0.560-0.786)	92.86 (0.805-0.985)	1.430	.658a
kμdiff (kPa)	0.818 (0.739-0.897)	69.44 (0.575-0.798)	92.86 (0.805-0.985)	5.155	1.000b
kμMRE (kPa)	0.818 (0.740-0.895)	91.67 (0.827-0.969)	57.14 (0.410-0.723)	5.781	.973c
csADC (10−3 mm2/s)	0.833 (0.757-0.909)	65.28 (0.531-0.761)	92.86 (0.805-0.985)	1.399	.882a
cμdiff (kPa)	0.834 (0.758-0.910)	63.89 (0.517-0.749)	92.86 (0.805-0.985)	5.038	.582b
cμMRE (kPa)	0.817 (0.740-0.894)	72.22 (0.604-0.821)	90.48 (0.774-0.973)	5.060	.597c
msADC (10−3 mm2/s)	0.813 (0.732-0.894)	65.28 (0.531-0.761)	85.71 (0.715-0.946)	1.389	.498a
mμdiff (kPa)	0.812 (0.730-0.893)	65.28 (0.531-0.761)	85.71 (0.715-0.946)	5.013	.123b
mμMRE (kPa)	0.724 (0.631-0.817)	68.06 (0.560-0.786)	71.43 (0.554-0.843)	5.167	.117c

Cutoff values were selected using the Youden index; values below the cutoff were classified as CKD or a higher stage. DeLong’s test was applied to evaluate and compare the diagnostic performance of sADC, μ_diff, and μ_MRE in detecting different degrees of renal dysfunction in CKD. P values are from DeLong’s test for pairwise AUC comparisons (Bonferroni-adjusted).

^asADC vs. μ_diff.

^bμ_diff vs. μ_MRE.

^csADC vs. μ_MRE.

Abbreviations: AUC, area under the curve; c, renal cortex; CI, confidence interval; CKD, chronic kidney disease; CKD_1-2, CKD stage 1-2; CKD_3-5, CKD stage 3-5; HV, healthy volunteers; k, kidney parenchyma; m, renal medulla; sADC (10⁻³ mm²/s), shifted apparent diffusion coefficient; μ_diff (kPa), diffusion-derived virtual shear modulus; µ_MRE (kPa), MR elastographic shear modulus.

Table 4.

Diagnostic performance of sADC, µ_diff, and µ_MRE in distinguishing mild from moderate-to-severe renal interstitial fibrosis in CKD.

	AUC (95% CI)	Sensitivity% (95% CI)	Specificity% (95% CI)	Cutoff value	P
ksADC (10−3 mm2/s)	0.814 (0.682-0.945)	78.57 (0.590-0.917)	78.26 (0.563-0.925)	1.491	.835a
kμdiff (kPa)	0.817 (0.685-0.948)	89.29 (0.718-0.977)	69.57 (0.471-0.868)	5.332	1.000b
kμMRE (kPa)	0.817 (0.693-0.940)	82.14 (0.631-0.939)	78.26 (0.563-0.925)	5.346	.967c
csADC (10−3 mm2/s)	0.910 (0.822-0.998)	85.71 (0.673-0.960)	91.30 (0.720-0.989)	1.444	1.000a
cμdiff (kPa)	0.910 (0.821-0.999)	85.71 (0.673-0.960)	91.30 (0.720-0.989)	5.173	.824b
cμMRE (kPa)	0.901 (0.812-0.989)	92.86 (0.765-0.991)	78.26 (0.563-0.925)	5.186	.816c
msADC (10−3 mm2/s)	0.849 (0.738-0.961)	75.00 (0.551-0.893)	91.30 (0.720-0.989)	1.413	1.000a
mμdiff (kPa)	0.849 (0.738-0.961)	75.00 (0.551-0.893)	91.30 (0.720-0.989)	5.083	.757b
mμMRE (kPa)	0.871 (0.771-0.971)	64.29 (0.441-0.814)	95.65 (0.781-0.999)	4.927	.757c

Cutoff values were selected using the Youden index; values below the cutoff were classified as moderate-to-severe interstitial fibrosis. DeLong’s test was used to compare AUCs of sADC, μ_diff, and µ_MRE. P values reflect pairwise comparisons (Bonferroni-adjusted).

^asADC vs. μ_diff.

^bμ_diff vs. μ_MRE.

^csADC vs. μ_MRE.

Abbreviations: AUC, area under the curve; c, renal cortex; CKD, chronic kidney disease; 95% CI, 95% confidence interval; k, kidney parenchyma; m, renal medulla; sADC (10⁻³ mm²/s), shifted apparent diffusion coefficient; μ_diff (kPa), diffusion-derived virtual shear modulus; μ_MRE (kPa), MR elastographic shear modulus.

Discussion

In this study, we developed a diffusion-based vMRE framework for assessing stiffness-related renal tissue alterations in CKD. By optimizing b-value combinations and modeling the relationship between sADC and MRE-derived shear modulus (μ_MRE), we demonstrated that vMRE achieved diagnostic performance comparable to MRE for assessing CKD-related structural alterations, while showing stronger correlations with renal function indices, supporting diffusion-based assessment of renal tissue properties related to stiffness without external mechanical vibration.

The optimal b-value combination for renal vMRE was 200-800 s/mm², with csADC_200-800 showing the strongest correlation with cμ_MRE (R² = 0.94, P < .001). This combination likely captures non-Gaussian diffusion effects in the 600-800 s/mm² range, which are more sensitive to renal microstructural alterations,²⁴ while avoiding signal loss and motion artifacts associated with ultra-high b-values.^25,26 Based on this optimized diffusion weighting, a CKD-specific μ_diff equation was derived: ${\mathrm{\mu }}_{\mathrm{diff}} \left(\mathrm{kPa}\right)=4.91 \ln (S_{200}/S_{800})+0.92 .$ Importantly, μ_diff should be interpreted as a diffusion-derived surrogate metric mapped onto a shear-modulus-like scale through empirical cross-modality calibration, rather than a direct mechanical measurement of tissue stiffness.²⁷

After adjustment for age, sex, and BMI, vMRE parameters (sADC, μ_diff) correlated more strongly with renal function indicators (eGFR, Scr) than MRE-derived μ_MRE, suggesting greater sensitivity to renal dysfunction. This differs from prior studies reporting stronger correlations between MRE-derived SWS and eGFR (r = 0.719 and 0.66, P < .05).^28,29 In our study, cortical and parenchymal μ_MRE showed weaker correlations with eGFR (r = 0.585 and 0.488, P < .001). This discrepancy may reflect the broader sensitivity of DWI-based vMRE to microstructural and stiffness-related alterations, such as TA and IF, differences in MRE acquisition parameters, particularly magnetic field strength,¹⁹ and cohort etiology. Hua et al.¹³ reported a stronger correlation between csADC and eGFR (r = 0.72, P < .001) than ours (r = 0.596, P < .001), possibly due to differences in ROI strategy. Their analysis used a single-slice ROI, whereas our multi-slice approach better captures renal parenchymal heterogeneity.³⁰ In both vMRE and MRE, cortical parameters showed stronger correlations with renal function than medullary parameters, supporting the rationale for cortical-based μ_diff model derivation.

When comparing group differentiation, vMRE and MRE demonstrated overall comparable performance, except that mμ_MRE did not significantly differentiate HV from CKD_1-2 (P > .05). This was further supported by ROC analysis, where mμ_MRE showed limited discrimination (P > .05). Although the diagnostic performance of vMRE was moderate, these findings support its potential as a noninvasive approach for assessing CKD-related changes, and further validation in larger cohorts is required. Both μ_diff and μ_MRE decreased progressively with CKD severity, contrasting with chronic liver disease, where stiffness increases with fibrosis progression,⁵ indicating that hepatic vMRE models are not directly transferable to native kidneys. Interestingly, renal allografts may exhibit fibrosis-related stiffness elevation due to transplant-specific pathophysiology, including ischemia-reperfusion injury and immune-mediated remodeling, suggesting that native-kidney μ_diff calibration may require modification for renal allografts.^31,32

Histologic validation supported the ability of vMRE to characterize CKD-related tissue alterations. Interstitial fibrosis is a hallmark of CKD progression, and tissue stiffness is a key diagnostic marker.³³ Magnetic resonance elastography directly quantifies stiffness and serves as a reference, whereas sADC reflects diffusion restriction without direct mechanical measurement. By incorporating μ_diff, diffusion-derived signals can be mapped onto a stiffness-related scale through empirical calibration, enabling stiffness-related assessment while preserving sensitivity to diffusion changes. Correlation analysis showed that μ_MRE correlated well with pathological indices, particularly GS (r_max = −0.550, P < .001), whereas vMRE parameters (sADC, μ_diff) showed stronger correlations with overall pathological score, IF, TA, and CV, suggesting greater sensitivity to microstructural and hemodynamic alterations. This difference may reflect their biophysical bases: MRE quantifies macroscopic stiffness via shear waves and is more responsive to extracellular matrix accumulation and glomerular changes,³⁴ whereas vMRE incorporates diffusion effects and captures finer structural heterogeneity. Accordingly, vMRE and MRE provide complementary information in CKD. For pathological grading, vMRE and MRE showed comparable performance, with cμ_diff and cμ_MRE achieving AUC > 0.9, without a significant difference.

Several limitations should be acknowledged. First, this single-center study used a single MRI system and lacked longitudinal follow-up; multicenter longitudinal studies are needed. Second, the μ_diff model was derived and validated within the same CKD_1-3 cohort without independent external validation, which may introduce overfitting; its applicability to CKD_4-5 remains uncertain and requires validation. Third, CKD staging was dichotomized (stages 1-2 vs. 3-5), potentially grouping patients with heterogeneous renal function and histopathologic features. Fourth, selection bias may have been introduced by excluding patients with poor image quality, and MRI-based ROIs were not spatially matched to biopsy sites. Although ROIs were delineated bilaterally on consecutive slices to provide a representative whole-organ assessment, intrarenal heterogeneity may still affect variability in radiologic-pathologic correlations, despite the predominance of diffuse renal diseases such as IgA nephropathy.³⁵ In addition, the CKD cohort included patients with heterogeneous underlying etiologies, and etiology-specific subgroup analyses were not performed, which may limit clinical applicability. Finally, age differences between HV and CKD groups were adjusted using ANCOVA; the cohort lacked ethnic diversity, and validation in age-matched and more diverse populations is warranted.

Overall, vMRE provides a noninvasive, vibration-free approach for evaluating stiffness-related renal microstructural alterations in CKD, with diagnostic performance comparable to conventional MRE. Estimated glomerular filtration rate reflects global renal function but is insensitive to early structural changes, whereas renal biopsy, although the reference standard, is invasive and unsuitable for longitudinal monitoring. In contrast, vMRE, derived from routine DWI, requires no additional hardware and provides a repeatable, accessible approach suitable for longitudinal follow-up and treatment monitoring. For patients requiring post-biopsy monitoring or those who poorly tolerate conventional MRE, such as patients with polycystic kidney disease, vMRE may serve as an adjunct imaging-based tool. Although this was a single-center study without external validation, the significant correlations between vMRE and histopathologic indices support further investigation and clinical translation. Future multicenter studies with larger cohorts are warranted to confirm robustness and reproducibility across populations, scanners, and imaging platforms. If externally validated, vMRE may offer a clinically applicable and scalable approach for stiffness-related renal tissue assessment.

List of contributors

Ruirui Qi and Junfeng Li contributed equally as joint first authors. Youjia Zeng and Yueyao Chen are co-senior authors.

Author contributions

Ruirui Qi (Conceptualization, Data curation, Formal analysis, Investigation, Project administration, Visualization, Writing—original draft), Junfeng Li (Conceptualization, Data curation, Formal analysis, Investigation, Project administration, Visualization, Writing—original draft), Wenxi Liu (Data curation, Formal analysis, Software), Peiyin Luo (Data curation, Investigation, Project administration, Resources), Qiuyi Chen (Formal analysis, Visualization), Qiumei Liang (Investigation, Methodology, Project administration), Fanqi Meng (Data curation, Software), Jinhua Qin (Data curation, Software), Feifei Qu (Methodology, Resources, Writing—review & editing), Haodong Qin (Methodology, Software), Yanglei Wu (Software), Hanqing Lyu (Resources, Supervision), Bingsheng Huang (Methodology), Youjia Zeng (Conceptualization, Resources, Supervision, Writing—review & editing), and Yueyao Chen (Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing—review & editing)

Supplementary material

Supplementary material is available at Radiology Advances online.

Funding

This work was supported by the Science and Technology Planning Project of Shenzhen Municipality, China (grant no. JCYJ20230807094602006) and the Guangdong Medical Research Foundation, China (grant no. A2024660).

Conflicts of interest

Please see ICMJE form(s) for author conflicts of interest. These have been provided as supplementary materials. The authors declare no conflicts of interest or competing financial or personal relationships associated with this work.

Data availability

Data generated or analyzed during the study are available from the corresponding author by request.