Chronic Kidney Disease (CKD), defined as kidney injury and/or loss of kidney function is associated with the apparition of renal fibrosis¹. Diagnostic tools and non-invasive biomarkers for the detection of renal fibrosis are essential to complement serologic markers and biopsies in order to improve the prognostic and follow-up of patients with renal diseases. Currently, there is no recognized non-invasive method to assess renal fibrosis. Multi-parametric studies are beginning to emerge in renal disease assessment^2-4. However these studies investigated each MR parameter independently comparing the MR sequences but do not combine multiple parameters in a single statistic. Multiple Linear Regression (MLR) attempted to model the relationship between several explanatory variables and a response variables⁵. Several MRI sequences are emerging to measure fibrosis, including T1 mapping^6,7 and Diffusion-Weighted MRI (DWI)^8,9 as the two most promising methods. These two sequences were not yet compared to assess renal fibrosis. In this multi-parametric MR study, the sensitivity of T1 mapping¹⁰ and readout-segmented DWI (RESOLVE¹¹) was first independently evaluated and compared against interstitial fibrosis of CKD undergoing renal biopsy. The two MR parameters were then associated in a single statistic with the hypothesis that used together the detection of interstitial fibrosis can be improved.

31 CKD patients scheduled for biopsy (characteristics in figure 1) were scanned at 3T on a MR Siemens Magnetom Trio Tim system. Sequence parameters for Modified Look-Locker Inversion recovery (MOLLI) T1 mapping and Readout Segmentation Of Long Variable Echo train (RESOLVE) DWI are summarized in figure 2. Regions-Of-Interest were placed for analysis of both the cortex and medulla and, ΔT1 and ΔADC were defined respectively as the difference between the cortical and medullary T1 and ADC. Interstitial fibrosis was quantified on Masson trichrome stained sections from biopsy. Pearson’s correlations between ΔT1, ΔADC and interstitial fibrosis were carried out. Correlation coefficient comparison was performed using Fisher Z-transform with strong correlations when p<0.05. MLR compared ΔT1 and ΔADC to the extent of renal fibrosis. In our study, MLR, given the relationship between χ1 and χ2 explanatory independent variables and yi dependent variables, is modeled by $$y_{i}=\beta_{0}+\beta_{1}x_{1}+\beta_{2}x_{2}+\epsilon$$ where β are the regression coefficient and ε is the residual error. We calculated R² and adjusted R² defined as $$R_{adjusted}^2=1-\frac{n-1}{n-m-1}(1-R^{2})$$ with n the number of observations and m the number of explanatory variables. The independence of explanatory variables and normal distribution with zero mean were verified with the fit residues to assure the robustness of the fit. Interobserver agreement for the ADC values measured in the cortex and medulla, as well as ΔADC, was also performed using two independent observers. 10 CKD were chosen randomly and inter-observer reproducibility was calculated using Pearson’s correlations and Intra-class Correlation Coefficient (ICC) using one-way random single measures in SPSS software.

The RESOLVE image quality was good, with only a few susceptibility artifacts at the edge of the parenchyma (figure 3). A moderate positive correlation was measured between ΔT1 and fibrosis (R²=0.29 p<0.001) and a strong negative correlation was measured between ΔADC and fibrosis (R²=0.68 p<0.001). Based on R² correlation comparison using the Fisher Z-transform test, ΔADC outperformed ΔT1 to assess interstitial fibrosis in CKD (p=0.068). The correlation between MR parameters and fibrosis was improved when comparing the both MR parameters together against the percentage of fibrosis (R²_adjusted=0.74, p<0.001). An increase of ΔT1 and a decrease of ΔADC were measured with the increase of fibrosis according to the slope coordinate of the correlation plan with 0.12 for ΔT1and -0.138 for the ΔADC (figure 4). The Q-Q plot confirmed that the distribution was compatible with a normal distribution as all the points formed a straight line diagonal. Also, points are equally distributed in space, meaning that the distance between the residues and the model was equally distributed (figure 5). Strong reproducibility of ADC measurement in the cortex and medulla was demonstrated between the two readers with ICCs for each patient independently for ADC cortex, medulla and ΔADC all > 0.91 [95% CI: 0.92-0.99]. Correlation coefficients between the two readers were R²=0.96 for the ADC evaluation in the cortex, R²=0.97 in the medulla and R²=0.95 for the ΔADC (p<0.05).RESOLVE yielded DWI of high quality in CKD¹². Although DWI obtained with RESOLVE sequence outperformed T1 mapping to assess renal fibrosis, the combination of both ΔADC and ΔT1 was better to assess fibrosis than either ADC or T1 alone.