Multiparametric MRI for Predicting Renal Function Deterioration and Chronic Kidney Disease Development in Patients Undergoing Nephrectomy for Renal Masses: A Pilot Study

Abstract

BACKGROUND:

Patients with solid renal masses (SRMs) are at risk of development and progression of chronic kidney disease (CKD) after surgical resection with no reliable pre-operative predictor.

PURPOSE:

To investigate whether pre-operative multiparametric MRI can predict CKD development and progression to stage 3 CKD.

STUDY TYPE:

Prospective.

POPULATION:

43 participants (female = 13, mean age: 59±12 years) undergoing nephrectomy for SRM.

FIELD STRENGTH/SEQUENCE:

1.5T, MRI with 9-b-value diffusion-weighted echo-planar-imaging (DWI), 3D spoiled-gradient-echo variable flip angle (T₁-mapping), multi-echo gradient-echo blood-oxygen-level-dependent (BOLD), and dynamic-contrast-enhanced 3D T₁-weighted gradient-echo MRI (DCE-MRI).

ASSESSMENT:

A clinical CKD risk score was calculated from estimated glomerular filtration rate (eGFR), age, diabetes, and surgery (partial or radical nephrectomy). Pre-operative MRI was evaluated for eGFR, prediction of post-operative eGFR-decline>5ml/min/1.73m², and prediction of stage 3 CKD development (eGFR<60 ml/min/1.73m²) at 12-month follow-up. Parameters included cortical and medullary apparent diffusion coefficient (ADC), intravoxel incoherent motion (IVIM), tri-exponential diffusion (fast, medium, and slow) and spectral diffusion (vascular, tubule, and tissue) from DWI-MRI, native T₁ from T₁-mapping, R₂* from BOLD, and renal plasma flow and eGFR from DCE-MRI.

STATISTICAL TESTS:

Mann-Whitney U-test and Spearman’s rank correlation coefficient (r). Diagnostic ability was determined by leave-one-out cross-validated logistic regression area-under-the-receiver-operator-curve (AUC) and diagnostic odds ratio (OR) with p-value<0.05 considered significant.

RESULTS:

Thirty (67%) participants had normal renal function (eGFR ≥60 ml/min/1.73m²) prior to resection. Of 29 participants (67%) who completed 12-month follow-up, 19 (66%) had baseline normal eGFR, and 7 (37%) developed stage 3 CKD. eGFR from DCE-MRI and tubule diffusion correlated with baseline eGFR ($r^2=0.43\text{and}0.33$ respectively). Reduced vascular diffusion predicted eGFR decline (AUC=0.75–0.83, OR=6.8–16.5). A larger contralateral ADC corticomedullary difference (AUC=0.89; OR=22.5), and clinical CKD risk score (AUC=0.81; OR=5.5) were the strongest predictors of CKD development.

DATA CONCLUSION:

Pre-operative MRI predicted post-nephrectomy CKD development. A larger corticomedullary difference in ADC may indicate reduced functional reserve.

plain language summary:

Participants with solid kidney tumors often undergo invasive surgery to remove the tumor without knowing their individual risk of developing chronic kidney disease afterwards. This study tested if multiparametric kidney MRI performed before surgery could predict decline in kidney function and progression of chronic kidney disease within a year. The results suggest that MRI provides complementary information beyond standard clinical and laboratory measures for predicting kidney disease. As new medications become available to protect kidney function, MRI combined with clinical and lab data may help guide treatment planning, monitoring, and early intervention for participants preparing to undergo kidney tumor surgery.

INTRODUCTION

Chronic kidney disease (CKD) is highly prevalent, affecting approximately 14% of US adults (35.5 million) in 2023(1). Early recognition is key to preventing or delaying progression to end-stage kidney disease(1). Patients undergoing partial or radical nephrectomy to remove solid renal masses (SRM) are at risk of CKD development or progression after surgery, even when preoperative renal function is normal(2, 3). Although several clinical and surgical predictors—such as age, diabetes, baseline estimated glomerular filtration rate (eGFR), and total kidney volume—have been proposed(4–6), there is still no clinically accepted method to identify which patients will develop CKD following nephrectomy.

Earlier detection of CKD, specifically stage 3 (eGFR<60ml/min/1.73m²), improves management and reduces subsequent decline in function, whereas diagnostic delay is linked to worse outcomes(7). Serum creatinine–based eGFR is widely available and inexpensive, but has limited ability to predict future deterioration or recovery(6), cannot capture single kidney function, and does not directly reflect underlying pathophysiology or compensatory changes(8). Consequently, risk assessment is often supplemented by invasive biopsy and imaging surveillance(9). Non-invasive kidney MRI may help predict renal function deterioration and assess single kidney function in patients scheduled for nephrectomy supporting personalized treatment planning for SRMs and earlier intervention for post-operative CKD.

MRI techniques including apparent diffusion coefficient (ADC), T1-mapping, blood-oxygen level-dependent (BOLD)/R₂*, and dynamic contrast enhanced MRI (DCE-MRI) have shown promise for assessing and predicting renal function(10–12): studies have found that reduced perfusion, elevated T₁, and reduced oxygenation are associated with CKD progression(13–15). However, studies specifically focusing on CKD post-nephrectomy are lacking.

Thus, the aims of this prospective study were to assess the role of pre-operative mpMRI and clinical measures for: 1) prediction of renal function deterioration within 12 months, 2) prediction of progression to stage 3 CKD and 3) assessment of split kidney function and baseline CKD in patients undergoing surgical management of SRMs.

METHODS AND MATERIALS

Participants

This prospective single-center study was approved by the Institutional Review Board of the Icahn School of Medicine at Mount Sinai and all participants provided written informed consent. Forty-three adults scheduled to undergo partial or radical nephrectomy for a localized solid renal mass at the Icahn School of Medicine from October 2018 to April 2023 were recruited (Figure 1). Exclusion criteria were age <18y, advanced stage (T4) or metastatic renal cancer, non-resectable renal masses, masses on transplanted kidneys, contra-indications to MRI, and pre-existing medical conditions including severe claustrophobia or seizure. Participants with eGFR< 30 ml/min/1.73m² did not undergo DCE-MRI. Participants underwent pre-operative MRI including multiparametric MRI sequences within 31 days before nephrectomy. Participant age, sex, race/ethnicity, and weight/body mass index (BMI) were recorded (Table 1). A subset of these participants has been included in previous studies on renal mass radiomics(16) and on characterization of renal mass pathology and immunophenotype (17); the current study is beyond the scope of these previous publications.

Figure 1.

Clinical flow chart.

Table 1.

Patient demographics and clinical characteristics

Demographics and Clinical Features
Sex(M/F)	30/13
Race
White/Caucasian	31
Black/African American	10
Other	2
Age (years; mean±std, range)	59±12 (range 33 – 86)
Weight (kg; mean±std, range)	87.6±21.1 (range 46.3 – 137.0)
BMI (kg/m2; mean±std, range)	29.4±5.8 (range 18.1 – 43.8)
Total kidney volume (ml; mean±std, range)	376±96 (range 206–643)
Tumor volume (ml)	34.5±48.1 (range 1.1 – 251.4)
Surgery Type (Partial/Radical)	5/45
Tumor subtype* (number of patients)
Clear Cell	22
Papillary	4
Chromophobe	1
Chromophobe w/ Oncocytoma	1
Clear cell papillary	3
AML	4
Oncocytoma	5
Cyst	1
Malignancy (Malignant/Benign)*	31/9
Grade (Number of patients)
Grade 1	10
Grade 2	17
Grade 3	2
Grade 4	2
TNM Stage AJCC 8th addition (Number of patients)
pT1a	19
pT1b	9
pT3a	3
Tumor size on histopathology (cm)	3.3±1.7 (range 0.8–8.0)
Baseline eGFR (CKD-EPI 2021 mL/min/1.73m 2 ; mean±std, range)	70.4±19.4 (range 37.3–130.5)
Baseline CKD stage
Stage 1 (≥90 mL/min/1.73m2)	5
Stage 2 (60–89 mL/min/1.73m2)	25
Stage 3 (30–59 mL/min/1.73m2)	13
Stage 4 (15–29 mL/min/1.73m2)	0
Stage 5 (<15 mL/min/1.73m2)	0
Proteinuria**
0	27
+1, +2, +3	5
Hematuria**
≤25%	29
>25%	5
History of Diabetes (y/n)	33/9
History of Hypertension (y/n)	28/14

^*Two patients included in this study at baseline did not follow through with scheduled surgery at MSH, and one had a cyst, so tumor histopathological classification is not available.

^**Baseline urinalysis and blood test results were available in clinical charts within 6 weeks of the MRI for 34 patients.

Clinical Biomarkers

Histopathologic tissue confirmation of tumor subtype of each lesion was extracted from the surgical pathology report after nephrectomy, and malignancy and grade were also recorded. In addition, history of hypertension and/or diabetes was collected from medical records. Proteinuria (1+, 2+, or 3+; >30 mg/dL) and hematuria (>25% red blood cells) were collected from clinical charts within 6 weeks of the MRI. eGFR was estimated at baseline, at 3-months post-operatively, and at 1 year post-operatively from serum creatinine using the CKD-EPI 2021 criteria(18). A clinical CKD risk score was generated from eGFR, age, diabetes status, and surgery technique following the simplified calculation proposed by Ellis et al. (4) developed for predicting CKD after nephrectomy (+1 point for age > 65yr, +1 point for history of diabetes, +3 points for radical nephrectomy, and +3 for eGFR <90ml/min/1.73m², with an added +1 if eGFR <80, and another added +1 if eGFR <70).

Image Acquisition

All participants were imaged after a 4-hour fast from food and liquids in a 1.5T MRI scanner (Magnetom Aera, Siemens Healthineers, Erlangen, Germany) with a 70 cm wide bore, 137 cm length, an 18-channel flexible body array and 32-channel integrated spine array coils. The protocol included multi-b-value DWI, BOLD/R₂*, and T₁-mapping with variable flip angle. The sequence acquisition parameters are summarized in detail in Supplement 1. Respiratory gating and on-scanner motion correction were used to reduce motion artifacts. A subset of 38 participants underwent DCE-MRI. A smaller subset of 28 participants underwent arterial spin labeling (ASL) perfusion imaging (Supplement 2) after the prototype ASL sequence became available on our scanner. Two participants had scans run twice in the same day, with the participant removed from the scanner between acquisitions, for test-retest repeatability of MR parameters of the kidney tissue.

Image Processing

All sequences were post-processed in-house (MATLAB R2023a, MathWorks Inc.), except ASL which was processed on-scanner, with examples shown in Figure 2. Detailed model types, fitting algorithms, and image parameters are included in Supplement 3 for T₁, BOLD/R₂*, ASL, and DCE-MRI, as well as multi-b-value DWI with equation models for apparent diffusion coefficient (ADC), intravoxel incoherent motion (IVIM), triexponential diffusion, and spectral diffusion. T₁ and R₂* were analyzed with conventional monoexponential models, while DCE-MRI perfusion, multi-component diffusion models, and ASL were analyzed as follows:

Figure 2.

Example MR images of contralateral and ipsilateral kidneys in 1) a patient who maintained healthy renal function (75yo, female, BMI=18.1, baseline eGFR=87.0, underwent partial nephrectomy for a 0.8 cm angiomyolipoma in the right kidney), 2) a patient who developed stage 3 CKD (46yo, male, BMI=30.9, baseline eGFR=63.2ml/min/1.73m², hypertension, underwent partial nephrectomy for a 3.6 cm oncocytoma in the left kidney) and 3) a patient (64yo, male, BMI=31.7, baseline stage 3 CKD with eGFR=45.7ml/min/1.73m2, hypertension, underwent partial nephrectomy for a 4.8cm WHO grade 2 clear cell carcinoma in the left kidney). Shown for each patient are (A) T1, (B) R2*/BOLD map in s⁻¹, (C) early phase DCE-MRI and (D) late DCE-MRI, and (E) DWI at b=10 and (F) DWI at b=400. Included for patient 3 are example parameter fits for (G) DCE gadolinium concentration curve (RPF: renal plasma flow, Hct: hematocrit, $Vt$ tissue volume, C_A: population-based arterial input function, R(t): impulse retention function) and (H) multi-b-value DWI (a.u.; arbitrary units). For multi-b-value DWI an example spectral diffusion spectrum is included with three peaks, corresponding to three exponentials that represent three distinct diffusion components.

• T₁: T₁ was fit voxel-wise from the slope of the first-order polynomial fit of the variable flip angles.

• BOLD-MRI: R₂* was extracted from the BOLD-MRI acquisition with a voxel-wise fit using non-linear least-squares estimation from 12 echo times.

• DCE-MRI: For DCE-MRI analysis, renal plasma flow (RPF) and eGFR were obtained by fitting a whole-kidney DCE-MRI model(19) using the Levenberg-Marquardt algorithm. In this approach, the population-based arterial input function (pAIF)(20) was convolved with an impulse retention function to match the renal concentration curves, yielding model parameters: filtration fraction (f) and renal plasma flow (RPF) (Figure 2G). The eGFR was then computed as their product (eGFR=RPF×f).

• ADC: ADC was fit voxel-wise with a least-squares log-linear fit of b=200, 400, and 800 s/mm².

• IVIM: Voxel-wise multi-b-value decay curves were fit to the conventional biexponential $\frac{S_b}{S_0}=f{e}^{-b{D}^{*}}+(1-f)e^{-bD}$ with a Bayesian estimation(21, 22). This study assumed log-priors for the kidney as log(D)=−6.2±1, and log(D*)=−3.5±1 in mm²/s and, by fitting to the biexponential, returned $f$ and $D^{*}$ representing capillary microcirculation, and $1-f$ and $D$ representing tissue diffusion(23, 24).

• Triexponential: Voxel-wise decay curves were fit to a triexponential with least-squares fitting, adding a third diffusion component(25). The starting values[bounds] of the three exponentials were $f_{\mathit{\text{slow}}}=0.7\left[\mathrm{0,1}\right]$, $D_{\mathit{\text{slow}}}=0.001\left[0,0.01\right]{\text{mm}}^2/\text{s}$, $f_{\mathit{\text{medium}}}=0.2\left[\mathrm{0,1}\right]$, $D_{\mathit{\text{medium}}}=0.01\left[0,0.1\right]{\text{mm}}^2/\text{s}$, and $f_{\mathit{\text{fast}}}=0.1\left[\mathrm{0,1}\right]$, $D_{\mathit{\text{fast}}}=0.1\left[0,0.5\right]{\text{mm}}^2/\text{s}$; after fitting, the exponentials were assigned in order of ascending $D$ from slow to fast. The product $fD$ of each compartment was also analyzed, e.g. ($f_{\mathit{\text{fast}}}\times D_{\mathit{\text{fast}}}=f{D}_{\mathit{\text{fast}}}$).

• Spectral Diffusion: The 9-b-value curves were fit to an unconstrained sum of exponentials with non-negative least squares (NNLS) of 300 logarithmically spaced values(24, 26, 27). This returned voxel-wise spectrums representing probability density distributions of diffusion coefficients (Figure 2H). The Tikhonov regularization parameter was set to 0.1 reduce computation time(24). Each spectral peak signal fraction ($f$) was calculated as the area under the peak while the corresponding diffusion coefficient ($D$) was calculated as the weighted average coefficient of the peak. The peaks were then sorted into either tissue, tubule, or vascular diffusion components by diffusion coefficient (Supplement 4). The product of each compartment fraction and diffusion coefficient was also analyzed (e.g. $f_{\mathit{\text{tissue}}}\times D_{\mathit{\text{tissue}}}=f{D}_{\mathit{\text{tissue}}}$); an overestimated signal fraction $f$ can be countered by an underestimated $D$, and vice versa, hence the combination may adjust for correlative errors(24).

• ASL: ASL renal blood flow was quantified with a single-compartment model(28).

Image Analysis

Observer 1 (SL, abdominal radiologist with 13 years of experience) measured pre-operative total kidney volumes, excluding tumors, and tumor volumes separately from T₁-weighted images using segmentation software (Vitrea, Vital Images, Inc., Minneapolis, MN, USA). For T₁, R₂*/BOLD, and multi-b-value DWI, circular regions-of-interest (ROIs) sampling the cortex and medulla were delineated on kidneys both ipsilateral and contralateral to SRM by Observer 1. These ROIs were drawn as 2 cortical and 2 medullary circular selections each of upper, middle, and lower pole on two hilar slices returning a total of 24 ROIs per kidney. ROIs were drawn on the raw images, rather than processed maps, to minimize bias. ROIs were analyzed first on a voxel-by-voxel basis and on an average ROI basis to compare between the two methods. In a subset of participants (n=27), ROIs were drawn on T₁, R₂*/BOLD and multi-b-value DWI by Observer 2 (AR, radiology resident with 2 years of experience) for measurement of interobserver reproducibility. Due to noise, ASL ROIs were drawn as whole-kidney ROIs excluding the lesion, rather than circular ROIs, by Observer 1, with ASL ROIs drawn by Observer 2 for interobserver reproducibility. Finally, to study if imaging characteristics of the tumor would be indicative of CKD, separate volumes-of-interest were placed encompassing the entire renal mass by Observer 1.

The DCE-MRI data was motion corrected by 4D autofocus coregistration between time points and a volumetric whole-kidney ROI (FireVoxel, NYU, USA) was drawn by Observer 3 (XM, research assistant with 3 years of experience), avoiding the tumor region. Observer 4 (OB, body MRI physicist with 11 years of experience) delineated whole-kidney ROIs on DCE-MRI in a subset of 10 participants for interobserver reliability.

Kidney MRI Parameters:

For T₁, R₂*/BOLD, ASL, ADC, and IVIM, histogram characteristics of the MR parameters included central tendency (mean, median) and standard deviation (stdev). Triexponential diffusion, spectral diffusion, and DCE-MRI used only central tendency parameters. All acquisitions were analyzed on a single-kidney basis (e.g. mean ipsilateral T₁ and mean contralateral T₁ were considered separate parameters). DCE-MRI eGFR was measured on a single-kidney basis and also combined for a DCE-MRI total eGFR. Ipsilateral and contralateral spectral diffusion $f{D}_{\mathit{\text{tubule}}}$, which has shown promise as a measure of function and fibrosis(24, 29), were combined for comparison to baseline eGFR and DCE-MRI total eGFR. Finally, corticomedullary difference was calculated as the difference between cortical ROIs and medullary ROIs for each of the central tendency parameters (e.g. mean cortical ADC – mean medullary ADC) with a “larger” corticomedullary difference indicating a difference that is further away from zero(30).

Tumor MRI parameters:

Central tendency and standard deviation of parameters from the volumetric ROIs of the renal masses were measured. As a tumor does not contain kidney tubules, multi-component diffusion was only analyzed with ADC and with IVIM which assumes tissue diffusion and capillary perfusion.

Study Endpoints

In this study, ‘CKD’ was defined as clinically significant stage 3 CKD and more severe stages (i.e. eGFR<60 ml/min/1.73m²). Overall renal function deterioration is a marker of CKD progression, and CKD stage 3 is associated with increased rates of renal failure and death (31). Cases with stage 2 or milder were considered ‘healthy’ in this study (eGFR ≥ 60 ml/min/1.73m²)(7). Accordingly, two primary endpoints were defined: (1) overall renal function deterioration and (2) development of CKD, defined as progression to stage 3 or higher.

Overall renal function deterioration, regardless of baseline, was defined as a decline in eGFR greater than 5 ml/min/1.73 m² over 12 months (baseline eGFR – 12-month eGFR > 5ml/min/1.73m²), a threshold that is associated with higher risk of end stage renal disease(32). The magnitude of renal function deterioration was also recorded as a continuous variable (change in eGFR within one year). Development of CKD was defined as progression to stage 3 CKD or higher, based on at least one measurement of eGFR<60 ml/min/1.73m² within one year in participants with a healthy baseline eGFR. As secondary endpoints, mpMRI was evaluated for characterization of baseline eGFR, split kidney function, and baseline stage 3 CKD.

Statistical Analysis

Differences of mean, median, and stdev of MRI-parameters at the voxel-wise level were assessed via the Mann-Whitney U-test (significance level p<0.05). For multiple comparisons corrections, the Benjamini-Hochberg procedure was applied with a false discovery rate of 0.35 to determine significance; this generous false positive rate was chosen to primarily minimize false negatives while allowing for exploration of novel MRI parameters. Univariable leave-one-out cross-validated logistic regression models were built to study each parameter’s classification ability independently. Multivariable models were also built with feature selection per loop to avoid data leakage; a maximum of two parameters were selected per loop. Success of an algorithm was calculated with area-under-the-receiver-operating-characteristic-curve and 95% confidence interval (AUC[95%CI]) via bootstrapping with sensitivity, specificity, and diagnostic odds ratio (OR) reported at the Youden’s J-statistic probability threshold. Models were compared with the DeLong’s test. Parameters with significant Mann-Whitney U-test results and significant cross-validated AUC are reported along with the group-level mean and SD of the parameter per group, the U-test p-value, and the AUC curve with 95% CI, sensitivity, specificity, the optimal threshold, and OR with corresponding p-value. These parameters with both Mann-Whitney U-test p<0.05 and cross-validated logistic regression p<0.05 were considered to have statistically significant predictive ability. Results for parameters of interest that did not meet these criteria are included in the supplement.

Correlation of mpMRI features to baseline CKD-EPI eGFR in mL/min/1.73m² was calculated with linear regression and Spearman’s rank correlation coefficient. Correlation of significant MR-parameters against the extent of renal deterioration was also calculated with Spearman’s rank correlation coefficient. Coefficient of variation (CoV) was used to examine test-retest repeatability of each MR parameter with CoV≤10% considered excellent, 10%-20% considered good, 20–30% considered acceptable, and CoV>30% considered poor(33). Interobserver reproducibility was calculated with intraclass correlation (ICC) returning reliability that is poor (ICC <0.50), moderate (0.5–0.75), good (0.75–0.90), and excellent (>0.90)(34). All statistical analysis was performed in Python 3.11.4 (Anaconda Inc., 2024).

RESULTS

Participant Demographics and Clinical Features

Forty-three participants (female = 13, mean age: 59 ± 12 years) scheduled to undergo nephrectomy and completed pre-operative MRI were included in this study (Table 1). The participant cohort included 13 participants with baseline CKD (30%), with the rate of stage 3 CKD increasing at 3-months (39%) and 12-months (48%) after surgery (Figure 1). Twelve-month follow-up eGFR measurement was available within a 30-day window for 29 (69%) participants, 19 (66%) of whom had baseline healthy kidney function. Seven (16.6%) participants developed CKD, and 12 (28.5%) participants maintained healthy function, while 14 (48%) had a serum creatinine CKD-EPI eGFR decline greater than 5 ml/min/1.7m². One case (healthy baseline, developed CKD) had incomplete multi-b-value DWI and was excluded from the DWI analysis.

Prediction of Renal Function Deterioration

Predictive MRI parameters determined by both significant Mann-Whitney U-test and significant cross-validated AUC for renal function deterioration (eGFR decline >5ml/min/1.73m² in one year) regardless of baseline eGFR, are shown in Table 2. ROC curves and corresponding boxplots of the imaging parameters with the highest AUCs are plotted in Figure 3. Multi-component diffusion MRI and T₁ parameters were significant predictors. Deterioration was predicted by reduced contralateral cortical parameters of vascular diffusion (triexponential median $f_{\mathit{\text{fast}}}$ AUC=0.83; OR=16.5; IVIM $f{D}^{*}$ AUC=0.79; OR=11.0) and reduced ipsilateral medulla parameters of the vascular diffusion components (triexponential $f{D}_{\mathit{\text{fast}}}$ AUC=0.78; OR=11.0; and spectral diffusion $f_{\mathit{\text{vasc}}}$ AUC=0.75; OR =6.75). The triexponential and spectral diffusion parameters also demonstrated significant monotonic correlation with the extent of deterioration (Table 2). Similarly, ipsilateral IVIM $f{D}^{*}$ was significantly lower in those in whom eGFR declined, but it did not return a significant cross-validated AUC or correlate with extent of decline (Supplement 5). Ipsilateral corticomedullary difference of $f{D}_{\mathit{\text{tubule}}}$ was significantly smaller in those who declined >5ml/min/1.73m² with an AUC=0.76, but did not significantly correlate with the extent of deterioration (r = 0.39, p=0.057). Elevated standard deviation of cortical T₁ significantly predicted decline (AUC=0.72; OR=5.87) without significant correlation (r=0.29, p=0.146) (Table 2). Ipsilateral DCE-MRI eGFR did return a significant correlation (r=0.47) and higher total DCE-MRI eGFR was a significant predictor of decline (AUC=0.75; OR=8.8). Cortical, medullary, and corticomedullary difference of R₂* in both ipsilateral and contralateral kidneys was not significant for prediction of decline in function (p=0.343–0.939; Supplement 5). R₂*, and ADC were not significant predictors, with at least either Mann-Whitney U-test or cross-validated AUC p≥0.05. Parameters of interest, regardless of significance, are included in Supplement 5 for reference. As ASL did not have a large enough sample size for predictive modeling, group level means and corresponding Mann-Whitney U test are included in Supplement 2, and none were significant.

Table 2.

a) Parameters with both Mann-Whitney U-test and AUC p<0.05 for prediction of >5ml/min/1.73m2 eGFR decline within one year of nephrectomy. Columns from left to right: name of parameter, the group-level mean and SD of the parameter per group, the U-test p-value, the cross-validated AUC value with 95% confidence interval, sensitivity (SN), specificity (SP), the optimal threshold, the diagnostic odds ratio (OR) with corresponding p-value, and the Spearman’s rank correlation coefficient between the parameter and the 1- year change in eGFR. Multiparametric models are provided in bold for each sequence, and were allowed to choose up to 2 features meaning it could choose only one variable. The kidney relative to the solid renal mass on which ROIs were drawn is contralateral (contra), ipsilateral (ipsi), and split into cortical and medullary territory. Diffusion coefficients units are [10e-3 mm²/s] unless indicated otherwise. A ♦ marks results that did not pass the Benjamini-Hochberg correction but had a raw p-value < 0.05. A * indicates a p<0.05. b) Conventional clinical features regardless of significance. A * indicates a p<0.05.

2a) Significant MRI Parameters	Decline ≤ 5 $\mu \pm \sigma$	Decline > 5 $\mu \pm \sigma$	U-test p-val	AUC[95%CI]	SN	SP	Youden’s J-stat threshold	OR; p-val	Spearman’s rank correlation coefficient $r,p$
IVIM	n=13	n=12		0.79[0.59, 1.0]*	0.75	0.85	0.445	11;* <0.001
median cortical $f{D}^{*}$ (contra)	5.91±2.43	3.93±0.97	0.008*	0.79[0.59, 1.0]*	0.75	0.85	0.445	16.5;* <0.001	r=−0.36, p=0.073
median cortical $D^{*}$ (contra)	23.6±6.0	17.7±5.9	0.012*	0.79[0.59, 1.0]*	0.67	0.85	0.445	11.0;* <0.001	r=-0.38, p=0.059
Triexponential	n=13	n=12		0.83[0.66, 1.0]*	0.75	0.85	0.501	16.5;* <0.001
median cortical $f_{\mathit{\text{fast}}}$ (contra)	0.11±0.05	0.06±0.03	0.003*	0.83[0.66, 1.0]*	0.75	0.85	0.501	16.5;* <0.001	r=−0.45, p=0.023*
mean medullary $f{D}_{\mathit{\text{fast}}}$ (ipsi)	20.4±3.7	15.5±5.1	0.014*	0.78[0.57, 0.98]*	0.67	0.85	0.617	11.0;* <0.001	r=−0.52, p=0.008*
corticomedullary difference of mean $f_{\mathit{\text{slow}}}$ (contra)	−0.03±0.05	0.03±0.05	0.019*	0.76[0.56, 0.96]*	0.42	1.0	0.504	2.2; 0.64	r=4.32, p=0.035*
mean cortical $f_{\mathit{\text{fast}}}$ (contra)	0.13±0.06	0.08±0.03	0.026*	0.72[0.53, 0.94]*	0.58	0.85	0.561	7.7;* 0.05	r=−0.323, p=0.115
Spectral	n=13	n=12		0.76[0.56, 0.95]*	0.75	0.62	0.502	4.8; 0.18
corticomedullary difference of median $f{D}_{\mathit{\text{tubule}}}$ (ipsi)	1.07±1.59	0.14±0.42	0.017*	0.76[0.56, 0.95]*	0.75	0.62	0.502	4.8; 0.18	r=−0.383 p=0.057
mean medullary $f_{\mathit{\text{vasc}}}$ (ipsi)	0.09±0.03	0.05±0.03	0.031♦	0.75[0.55, 0.95]*	0.75	0.69	0.500	6.75;* 0.05	r=−0.42, p=0.037*
T1	n=14	n=14		0.72[0.51, 0.93]*	0.62	0.79	0.521	5.87; 0.08
stdev cortical T1 (ipsi)	340±139	556±236	0.020*	0.72[0.51, 0.93]*	0.62	0.79	0.521	5.87; 0.08	r=0.29, p=0.150
DCE	n=13	n=13		0.70[0.47, 0.92]	0.69	0.69	0.482	5.06; 0.13
mean DCE eGFR (total)	54.54± 13.4	65.15± 7.83	0.020*	0.75[0.54, 0.96]*	0.85	0.62	0.408	8.8;* 0.02	r=0.38, p=0.056
2b) Clinical Features	n=15	n=14
Baseline CKD-EPI eGFR [ml/min/1.73m2]	65.5±15.6	77.8±13.5	0.034*	0.69[0.47, 0.91]	0.75	0.62	0.453	4.8; 0.18	r=0.41, p=0.044*
Creatinine (mg/dL)	1.1±0.2	0.84±0.16	0.003*	0.83[0.66, 0.99]*	0.75	0.77	0.508	10.0;*<0.001	r=−0.59, p=0.002*
Tumor size [ml]	52.2±69.6	41.2±39.1	0.828	0.54[0.23, 0.85]	0.50	0.50	0.480	1.0; 1.0	r=0.05, p=0.810
Total kidney volume [ml]	385±99	386±90	0.663	0.46[0.18, 0.73]	0.67	0.50	0.497	2.0; 0.81	r=−0.13, p=0.531
CKD Risk Score	5.33±1.58	4.43±2.35	0.275	0.54[0.31, 0.76]	0.14	0.93	0.588	2.3; 0.84	r=−0.19, p=0.320

Figure 3.

A) ROC curves of significant MR parameters (solid lines) and clinical parameters (dashed lines) for prediction of greater than 5ml/min/1.73m² drop in eGFR within 12 months of nephrectomy. An asterisk * marks those that have a significant AUC with a dotted grey line representing the line of random guessing. B) Corresponding boxplots with the medians, upper and lower quartiles, and whiskers are shown with corresponding Mann-Whitney U test p-values. A scatter plot of the individual data points, each representing a single patient, is superimposed with black data points representing those whose eGFR did not decrease by more than 5ml/min/1.73m² and red data points representing those whose eGFR did decrease by more than 5ml/min/1.73m².

From clinical features (Table 2), lower baseline serum creatinine predicted >5ml/min/1.73m² decline in renal function (AUC=0.83; OR=10.0) and negatively correlated with the extent of deterioration (r=−0.59); those with better pre-operative function were more likely to decrease in function after surgery. CKD-EPI eGFR, CKD clinical risk score, and age, BMI, history of hypertension, history of diabetes, total kidney volume, proteinuria, hematuria, tumor size, tumor malignancy, and histological subtype were not significant predictors of renal decline in this pilot study (p=0.275–0.973, Table 2, Supplement 5).

Prediction of Progression to Stage 3+ CKD

No participant progressed beyond stage 3 within the 12-month follow-up. Predictive MRI parameters determined by both significant Mann-Whitney U-test and significant cross-validated AUC for stage 3 CKD development are shown in Table 3, with corresponding ROC curves, boxplots, and representative cases shown in Figure 4. The contralateral ADC corticomedullary difference achieved the highest AUC of all metrics in this study (AUC=0.89; OR=22.5), consistent with the visually larger corticomedullary contrast in kidneys that subsequently developed CKD (Figure 4C), compared with moderate and minimal contrast in kidneys with preserved function and baseline CKD, respectively. Spectral diffusion MRI also predicted CKD, with reduced tissue diffusion (AUC=0.83; OR=13.3) and greater tubular diffusion in the ipsilateral medulla (AUC=0.79). The multivariable model using only mean cortical and medullary $f{D}_{\mathit{\text{tissue}}}$ was also significant (AUC=0.83). Ipsilateral T₁ corticomedullary difference was also significant (AUC=0.71). Higher ipsilateral cortical and medullary ADC achieved a significant cross-validated AUC for CKD but not a significant Mann-Whitney U-test result (Supplement 6). Cortical, medullary, and corticomedullary difference of R₂* in both ipsilateral and contralateral kidneys was not significant for prediction of development of stage 3 CKD (p=0.310–0.684; Supplement 6). DCE-MRI also returned no statistically significant parameters (p=0.162–0.841; Supplement 6). Parameters of interest, regardless of significance, are summarized in Supplement 6.

Table 3.

a) Parameters with both Mann-Whitney U-test and AUC p<0.05 for prediction of stage 3 CKD development from a healthy baseline within 12 months of nephrectomy. Columns from left to right: name of parameter, the group-level mean and SD of the parameter per group, the U-test p-value, the cross-validated AUC value with 95% confidence interval, sensitivity (SN), specificity (SP), the optimal threshold, and the diagnostic odds ratio (OR) with corresponding p-value. Multiparametric models are provided in bold for each sequence. The kidney relative to the solid renal mass on which ROIs were drawn is contralateral (contra), ipsilateral (ipsi), and split into cortical and medullary territory. Diffusion coefficients units are [10e-3 mm²/s] unless indicated otherwise. A ♦ marks those that did not pass the Benjamini-Hochberg correction but had a raw p-value < 0.05. The number of patients in each group is included for each sequence. A * indicates a p<0.05. b) Conventional clinical features regardless of significance. A * indicates a p<0.05.

3a) Significant MRI Parameters	Stable function $\mu \pm \sigma$	Developed CKD $\mu \pm \sigma$	U-test p-val	AUC[95%CI]	SN	SP	Youden’s J-stat threshold	OR; p-val
ADC [10−6 mm2/s]	n = 12	n = 6		0.89[0.71, 1.0]*	0.83	0.82	0.312	22.5; < 0.001*
corticomedullary difference of mean ADC (contra)	−21±94	167±86	0.002*	0.89[0.71, 1.0]*	0.83	0.82	0.312	22.5; < 0.001*
corticomedullary difference of median ADC (contra)	−18±73	173±84	0.002*	0.83[0.51, 1.0]*	0.83	0.82	0.501	22.5; <0.001*
Spectral Diffusion	n=12	n=6		0.77[0.53, 1.0]*	0.67	0.73	0.497	5.3; 0.33
mean medullary $f_{\mathit{\text{tubule}}}$ (ipsi)	0.54± 0.16	0.72± 0.11	0.016*	0.79[0.54, 1.0]*	0.50	0.91	0.512	10.0; 0.19
mean medullary $f{D}_{\mathit{\text{tissue}}}$ (ipsi)	0.40± 0.24	0.14± 0.09	0.027*	0.83[0.63, 1.0]*	0.83	0.73	0.501	13.3; 0.05*
mean medullary $D_{\mathit{\text{tissue}}}$ (ipsi)	0.91± 0.26	0.59± 0.17	0.027*	0.83[0.63, 1.0]*	0.83	0.73	0.489	13.3; 0.05*
median medullary $f_{\mathit{\text{tubule}}}$ (ipsi)	0.56±0.29	0.79±0.10	0.035*	0.74[0.51, 1.0]*	0.83	0.55	0.502	6.0; 0.41
mean medullary $f_{\mathit{\text{tissue}}}$ (ipsi)	0.39± 0.17	0.22± 0.10	0.035*	0.80[0.58, 1.0]*	0.83	0.64	0.490	8.8; 0.20
median medullar $f_{\mathit{\text{tissue}}}$ (ipsi)	0.37±0.28	0.12±0.09	0.044*	0.80[0.58, 1.0]*	0.83	0.64	0.490	8.8; 0.20
cortical and medullar mean $\mathit{f}{\mathit{D}}_{\mathit{\text{tissue}}}$ Model				0.83[0.63, 1.0]*	0.83	0.64	0.482	8.8; 0.20
T1	n=12	n=7		0.71[0.51, 1.0]*	0.50	0.92	0.691	11.0; 0.15
corticomedullary difference of median T1 (ipsi)	−387±207	−188±104	0.049*	0.71[0.51, 1.0]*	0.50	0.92	0.691	11.0; 0.15
3b) Clinical Features
Baseline CKD-EPI eGFR [ml/min/1.73m2]	83.3±7.8	72.2±8.6	0.025*	0.78[0.46, 1.0]	0.67	0.83	0.491	10.0; 0.07
Creatinine (g/mL)	0.87± 0.17	0.89± 0.12	0.963	0.33[0.07, 0.60]	0.67	0.25	0.499	0.7; 0.94
CKD Risk Score	3.33±1.49	5.17±1.46	0.028*	0.81[0.53, 1.0]*	0.33	0.92	0.638	5.5; 0.55
Tumor size [ml]	29.2±35.3	39.1±47.9	0.454	0.54[0.23, 0.85]	0.33	0.58	0.490	1.0; 0.94
Total kidney volume [ml]	381±128	377±42	0.779	0.46[0.18, 0.73]	0.33	0.25	0.500	0.83; 0.86

Figure 4.

A) ROC curve of significant MR parameters (solid lines) and clinical parameters (dashed lines) for prediction of development of CKD within 12 months after nephrectomy. An asterisk * marks those that have a significant cross-validated AUC. B) Corresponding boxplots with the medians, upper and lower quartiles, and whiskers are shown with corresponding Mann-Whitney U test p-values. A scatter plot of the individual data points, each representing a single patient, is superimposed with black data points representing those who did not develop CKD and red data points representing those who did develop CKD. C) Example ADC maps of the contralateral kidney scaled from 0 to 0.007 mm²/s for patients 1, 2, and 3 from Figure 1. This demonstrates ADC corticomedullary difference in patients who 1) maintained healthy function after nephrectomy 2) who developed stage 3 CKD within 12mo after nephrectomy, and 3) who had baseline stage 3 CKD.

For clinical features (Table 3), a higher CKD clinical risk score was associated with CKD development and showed acceptable discriminative performance (AUC = 0.81), though the OR was not significant (OR=5.50, p=0.55). The CKD clinical risk score model had high specificity but low sensitivity (specificity = 0.92, sensitivity = 0.33). Although lower mean baseline eGFR, older age, higher BMI, diabetes and hypertension were more frequent in participants who developed CKD, these parameters were not significant predictors of CKD, in addition to tumor volume, total kidney volume, proteinuria, serum creatinine, and hematuria (p= 0.092–0.779; Table 3, Supplement 6). Tumor characteristics of malignancy, and tumor histological subtype were also not significant predictors (p=0.291–0.704; Supplement 6).

Integrated Clinical-Imaging Models

For prediction of CKD development, the mpMRI model given the option to select two features from all sequences achieved an AUC(95%CI)=0.89(0.71,1.0), sensitivity=0.83, specificity=0.82, optimal threshold=0.321 with OR=22.5 based on larger contralateral ADC corticomedullary difference and elevated ipsilateral medullary spectral diffusion mean $f_{\mathit{\text{tubule}}}$. Addition of baseline eGFR or CKD clinical risk score to mpMRI models did not significantly improve the AUC (ADC corticomedullary difference + clinical risk score AUC(95%CI)=0.91(0.73, 1.0), OR=50.0, Delong p-value = 0.479; ADC corticomedullary difference + eGFR AUC(95%CI)=0.92(0.80, 1.0), OR=22.5, DeLong p-value = 0.646). In this pilot study, the mpMRI model showed AUCs comparable to or higher than clinical and integrated clinical-MRI models.

Imaging Split Kidney Function and Baseline CKD

DCE-MRI and spectral diffusion were analyzed both as split individual kidney parameters, and total kidney parameters (i.e. summed parameters from both kidneys). Ipsilateral, contralateral, and total DCE-MRI eGFR and spectral diffusion $f{D}_{\mathit{\text{tubule}}}$ significantly correlated with baseline CKD-EPI eGFR (Supplement 7). Total DCE-MRI eGFR showed the strongest correlation, and significant linear regression, to baseline CKD-EPI eGFR. Without radiotracers to measure true single-kidney GFR, the split kidney DCE-MRI eGFR and spectral diffusion $f{D}_{\mathit{\text{tubule}}}$ were examined relative to total CKD-EPI eGFR (Supplement 8). For single-kidney, the slopes of the linear regressions appeared approximately halved compared to the linear regression slopes of summed parameters for both kidneys, supporting DCE-MRI and spectral diffusion capturing signal from both kidneys contributing to total renal filtration (Supplement 8: spectral diffusion $f{D}_{\mathit{\text{tubule}}}$ ipsilateral slope = 0.06, contralateral slope = 0.06, total slope = 0.13; DCE-MRI eGFR ipsilateral slope = 0.24, contralateral slope = 0.24, total slope = 0.48).

While split kidney parameters demonstrated ability to predict CKD development with ipsilateral spectral diffusion, and both ADC and T₁ corticomedullary difference (Table 3), the combination of reduced total DCE-MRI eGFR and RPF was the best imaging biomarker of baseline CKD (Supplement 7: AUC=0.86; OR=8.57). Multi-b-value diffusion including ADC stdev, smaller ipsilateral IVIM $D$ corticomedullary difference, elevated ipsilateral IVIM $D^{*}$ and smaller T₁ corticomedullary difference were also significant for diagnosing baseline stage 3 CKD (Supplement 7). R₂* heterogeneity was significantly elevated in participants with baseline CKD but did not return a significant AUC (AUC=0.68, p=0.137) despite correlating significantly (Spearman’s r=−0.35) (Supplement 7). Similarly, the mean cortical R₂* in the contralateral kidney was lower in those with baseline CKD without a significant cross-validated AUC (AUC=0.67, p=0.095), although it did significantly correlate (Spearman’s r = 0.32) (Supplement 7). ASL renal blood flow measurements were not significant, although the mean ASL RBF was higher in the ipsilateral kidneys with baseline CKD (p = 0.325, Supplement 2).

Tumor MRI Parameters

No tumor parameter returned a significant cross-validated AUC for prediction of decline in function, prediction of stage 3 CKD development, or detection of baseline CKD (AUC p=0.054–0.963, Supplement 5, 6, 7). In this study the most relevant parameters for CKD detection and prediction were from characterization of the kidney parenchyma.

MRI Parameter Reliability

ICC and test-retest reliability results are shown in Supplement 9; two participants completed test-retest scans for CoV%. Central tendency parameters of R₂*, T₁, ADC, and IVIM $f$ and $D$ showed excellent test-retest reliability and good-to-excellent inter-observer agreement (CoV%=1.42–5.59; ICC=0.77–0.96). IVIM $f{D}^{*}{,D}^{*}$, triexponential, and spectral diffusion parameters showed acceptable-to-good CoV% (CoV% = 16.02±9.5) and moderate-to-good ICC on average (ICC =0.69±0.22).

DISCUSSION

Nephrectomy is currently the main treatment for renal cell carcinoma, and post-nephrectomy CKD increases chances of patient mortality and end-stage kidney disease(2). Predicting which patients undergoing surgical management of solid renal masses are at risk of developing or progressing CKD would allow clinicians to weigh CKD against other considerations when planning management of renal masses. In this study, pre-operative MRI, specifically 9-b-value DWI and T₁-mapping, predicted if a patient would develop stage 3 CKD after partial or radical nephrectomy and predicted overall renal function deterioration. Loss of kidney tissue after nephrectomy commonly leads to an acute reduction in eGFR(35) and a long-term reduction in renal function in a subset of patients(36). Given the chance of CKD development and renal decline post-nephrectomy, minimizing unnecessary surgery, knowledge of each kidney’s split renal function, and being able to identify patients at high risk of post-operative CKD with beyond just laboratory factors(4, 5) may help improve patient outcome(37). It could also inform treatment decisions, including the administration of medications including angiotensin-converting enzyme inhibitors, angiotensin II receptor blockers, and sodium-glucose transport 2 inhibitors(38) for prevention or delay of irreversible kidney damage. This study supports pre-operative MRI for prediction of CKD development after nephrectomy, and the need for future study of MRI in post-operative kidney disease.

Pre-operative multi-component DWI predicted functional decline, with lower vascular flow and perfusion correlating with greater future decline. Reduced cortical vascular parameters of both kidneys prior to surgery predicted post-operative decline in eGFR after nephrectomy. Elevated ipsilateral T₁ heterogeneity also predicted decline, potentially detecting heterogeneous inflammation or fibrosis as a disease marker. Counterintuitively, increased serum creatinine level was associated with a smaller decline; patients with CKD may be less likely to exhibit eGFR decline compared to those with normal kidney function(32). This was also observed with elevated DCE-MRI total eGFR predicting greater decline. The clinical CKD risk score was not a predictor of decline in function, as it was designed specifically for prediction of clinically significant stage 3 CKD post-nephrectomy(4).

The contralateral ADC corticomedullary difference achieved the best model for prediction of post-operative CKD development. Whereas smaller ADC corticomedullary difference has also shown utility for prediction of CKD in native kidneys with type 2 diabetes(39), this study supports a larger difference predicting post-operative CKD. A larger ADC corticomedullary difference in the contralateral kidney may signal that kidneys are at full capacity and therefore cannot compensate further after nephrectomy. Compensatory elevated cortical flow and tubular filtration may cause a faster signal decay in the b=200–600 s/mm² range; kidneys that exhibit hyperfiltration with an elevated ADC corticomedullary difference may have a reduced functional reserve, and kidneys with a reduced renal functional reserve may be at higher risk of developing stage 3 CKD after nephrectomy due to inability to compensate(40)(41).

Multi-component DWI and T₁ also predicted post-operative CKD. Whereas increased cortical T₁ itself did not predict CKD development or renal decline, contrary to recent studies of native and transplanted kidneys(12, 42), a smaller ipsilateral T₁ corticomedullary difference with decreased medullary T₁ and increased cortical T₁ was predictive. This may be due to prior studies not involving the impact of nephrectomy, and our pilot study sample size. Spectral diffusion also achieved significant cross-validated AUCs, though IVIM and triexponential diffusion did not. Ipsilateral spectral diffusion demonstrated increased tubular diffusion fraction in participants who developed CKD, supporting pre-operative hyperfiltration in the renal tubules predicting post-operative CKD observed with ADC. Further, renal tumor compressing the surrounding tissue may result in elevated flow in the ipsilateral parenchyma(43); kidney damage and loss of nephrons after resection then lead to decline in function. As no parameter from BOLD or DCE-MRI predicted future CKD, and eGFR itself also did not return significant AUC values, pending validation in a larger study, physiologic parameters such as hyperfiltration, microstructure, microvasculature, may be most indicative of post-operative CKD.

As the clinical CKD risk score proposed by Ellis et al. (4) demonstrated high specificity but low sensitivity, the success of multi-b-value DWI and T₁ MRI supports future work combining the information on patient-level health (age, diabetes status, eGFR, surgical technique) with pre-operative DWI or T₁-mapping that provides an in vivo snapshot of split kidney parenchyma pathophysiology. In addition, while the clinical CKD risk score performed well for prediction of stage 3 CKD, it did not predict eGFR decline. Co-morbidities of diabetes, hypertension, higher BMI, older age, and a larger malignant tumor were more common in patients who developed CKD after nephrectomy, but these indicators were not significant predictors of CKD development. This supports a previous study that found preoperative eGFR, age, sex, diabetes status, and hypertension were not indicative of CKD progression alone(44). As such, MRI may add complementary information to improve prediction of overall decline to enable earlier intervention.

Imaging allows splitting of kidney function into ipsilateral and contralateral kidneys which may aid personalized medicine. Total and split kidney DCE-MRI eGFR and spectral diffusion $f{D}_{\mathit{\text{tubule}}}$ correlated with baseline CKD-EPI eGFR, and decreased total DCE-MRI eGFR detected baseline stage 3 CKD with the highest AUC; DCE-MRI and spectral diffusion were sensitive to individual kidney function(45). Cortical R₂* in the contralateral kidney correlated positively with eGFR, but did not demonstrate higher hypoxia as a marker of baseline CKD(46, 47). IVIM returned higher ipsilateral medullary and cortical $D^{*}$ in baseline CKD agreeing with findings from a previous study in peritumoral kidney tissue(43). Larger tumors in patients with CKD at baseline likely demonstrate greater angiogenesis and lead to compression of renal tubules and vasculature that impair kidney function. This elevated ipsilateral IVIM $D^{*}$ was supported by elevated ASL RBF although direct comparison is limited due to the reduced ASL sample size and that IVIM measures ‘local’ flow and is independent of a delay, tagging plane, or arterial input function unlike unlike ASL or DCE-MRI(48, 49).

Knowledge of MRI parameter reproducibility is essential to validating these measures clinically. Interobserver agreement was excellent for whole-kidney DCE-MRI, and for R₂* and ADC. IVIM $f$ and $D$ were good-to-excellent with good CoV. Triexponential fast compartment parameters showed better interobserver agreement than the medium component and the slow component. Spectral diffusion showed better interobserver agreement than the triexponential fit, with all returning moderate-to-excellent reliability, except for median $D_{\mathit{\text{tubule}}}$, and with tubule and tissue parameters returning good-to-excellent reliability. The higher interobserver agreement of IVIM compared to triexponential or spectral diffusion may be due to the observed increased stability with a Bayesian algorithm(21, 24); however, IVIM returned fewer significant predictive parameters for CKD development than the ADC or spectral diffusion models.

Limitations

We recognize that the sample size of this study limits the generalizability of the results, with need for larger future studies and clinical trials. In addition, due to the routine clinical follow up of post-operative patients, CKD was defined by one value within 12 months rather than two measurements across three months. Further, to maintain suitable longitudinal information for prediction of CKD development, completed pre-operative MRI, eGFR, and 12-month follow-up eGFR along with restriction to patients who showed healthy function at baseline limited the number of patients that could be assessed for stage 3 CKD development. While total kidney volume was not predictive in this work, patients who developed CKD did have a larger tumor on average. Further work is needed on kidney volume post-nephrectomy as a predictor of CKD. With a larger dataset, machine learning models beyond logistic regression may also improve prediction ability, and mGFR rather than eGFR, as well as post-operative MRI are worthy of future study. In addition, T₁-mapping may benefit from B₁ inhomogeneity correction, and ASL required long scan time with respiratory tracking and was only acquired in a subset of patients. R2* was potentially confounded by tumor angiogenesis, inflammation, and perfusion variability. Further, multi-b-value DWI required a longer acquisition time with respiratory gating and lower spatial resolution and may benefit from more advanced fitting models(50). Improved SNR at 3T and whole-region ROIs may improve overall test-retest reliability and interobserver agreement.

Conclusions

Pre-operative mpMRI may provide complementary information to clinical measures for predicting CKD progression and functional deterioration post-nephrectomy and allows assessment of individual kidney function in patients undergoing surgical management of renal masses. With an increasing number of new medications with renal protective effects, classification and prediction of CKD progression and renal decline using pre-operative MRI as well as clinical risk factors have potential to enhance monitoring and early initiation of medication treatment, and to improve treatment planning in patients undergoing surgery and management of post-operative CKD.