The local ethics committee approved this prospective study, and all patients provided informed consent. Between March 2019 and July 2022, 43 patients who received a definitive clinical diagnosis of chronic liver disease and an imaging diagnosis of liver cirrhosis at our hospital were enrolled. Alanine transaminase (ALT), aspartate aminotransferase (AST), total bilirubin (TBil), blood urea nitrogen (BUN) and Scr were measured. The estimated glomerular filtration rate (eGFR) was calculated within 1 week before the MRI. Patients with liver cirrhosis were divided into two subgroups: cirrhosis with normal renal function (NR; the cirrhosis-NR group) and cirrhosis with RD (the cirrhosis-RD group). Cirrhosis-RD was diagnosed according to the criteria for hepatorenal syndrome-AKI (4).

Exclusion criteria were primary kidney diseases such as nephrotic syndrome and glomerular nephritis; hypertension, diabetic mellitus, or gout disease; serious systemic or organic lesions; contraindications for MRI; and use of medications such as diuretics, angiotensin-converting enzyme inhibitors, or angiotensin receptor blockers before MRI examination. Four patients with incomplete images and two with image artifacts were also excluded.

Twenty-three age and body mass index (BMI) matched healthy volunteers with no history of renal disease, diabetes, hypertension, or vascular disease were recruited as the normal control (NC) group. All NCs had normal BUN, Scr and eGFR levels and no renal lesions according to conventional T2-weighted MRI. Table 1 lists the demographic data for all participants.

All participants were required to fast for ≤6 h before MRI. All MRI examinations were performed on a 3T MR system (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany) equipped with an eighteen-channel body array coil. Conventional T2-weighted imaging-interpolated breath-hold examinations were performed to evaluate renal lesions. T1 mapping and ASL were used to quantify renal microstructure and perfusion changes. Table 2 lists the scanning parameters of all MRI sequences.

T1 mapping was performed using a 2D Look-Locker sequence. The sequence performed using a prototype inversion recovery snapshot FLASH acquisitions after a 180° inversion pulse, providing images at multiple inversion times (16 TIs at 104, 293, 482, 671, 860, 1049, 1238, 1427, 1616, 1805, 1994, 2183, 2372, 2561, 2750, and 2939 ms). Four slices centered on the renal hilum were obtained for each patient. The T1 parameter map was generated inline after data acquisition by the scanner, and was based on a compartment model processing images pixel-by-pixel.

ASL was obtained using a prototypical three-dimensional TGSE research sequence with a pseudo-continuous ASL (pCASL) scheme. Abdominal aorta was labeled with a labeling duration of 1500 ms and post-labeling delay of 3000 ms. One segment with 20 repetitions was performed. Perfusion map was estimated on a pixel basis using the following formula:

where f is the perfusion rate (in mL/100g/minute); λ represents the blood-tissue partition coefficient (0.9 mL/100 g); M ₀ is the equilibrium magnetization intensity; α is the inversion efficiency (0.98); Δ t is the arrival time of labeled blood (assumed to be 750 ms); T_1,blood is the longitudinal relaxation time of arterial blood; τ is the labeling time. Images were acquired in free breathing mode, and motion correction performed with motion compensation using 3D rigid and elastic registration algorithms.

Two experienced radiologists (with 11 and 3 years of experience each) who were blinded to the clinical data evaluated all MRI images. Renal injury in patients with liver cirrhosis is a systemic disease, main reason is the hemodynamic changes, so we assume the injury was similar between the two kidneys. MRI measurements were taken on bilateral kidneys separately. Three sections nearest the renal hilum were selected to draw the region of interest (ROI). For each selected section, one free-hand ROI (5–9 cm²) covering the entire renal cortex was manually drawn on both the T1 and RBF images, and three ellipsoid ROIs (0.2–0.4 cm²) were manually and separately placed on the upper, middle, and lower poles of the renal medulla on the T1 image.

Statistical analyses were performed using SPSS (version 25.0; IBM), MedCalc (version 19.0.4) and Python (version 3.9) software. Normally distributed variables are presented as means ± standard deviation and non-normally distributed variables are presented as medians (median, absolute deviation) after performing the Shapiro-Wilk test. The male/female ratio for each group was calculated via the chi-square test. Age, BMI, biochemical indices and renal MRI parameters for each group were compared using one-way analysis of variance (normal distribution and equal variances) or the Kruskal-Wallis test (non-normal distribution or equal variances). Post hoc multiple pairwise comparisons were performed using Bonferroni tests. Pearson/Spearman correlation tests were used to assess correlations between renal MRI parameters (cT1, mT1, and RBF) and biochemical indices (BUN, Scr, eGFR, ALT, AST and TBil). Receiver operating characteristic (ROC) curves were used to evaluate the diagnostic performances of the MRI parameters. Areas under the curve (AUCs) were compared using the DeLong method. A classification and regression tree (CART) analysis using cross-validation was performed using renal MRI parameters to identify renal impairment in patients with cirrhosis. CART analysis results are presented as a decision tree. Inter-reader reliability for measuring MRI imaging data was calculated using intraclass correlation coefficients (ICCs) and Bland-Altman analysis. p<0.05 was considered statistically significant.

Twenty-three healthy volunteers (NCs), 27 patients with cirrhosis and NR, and 10 patients with cirrhosis and RD were included in this study ( Table 1 ). Sex, age, BMI and ALT did not significantly differ among the three groups (all p>0.05). BUN, Scr, and eGFR did not significantly differ between the NCs and cirrhosis-NR group (all p>0.05) but were significantly higher in the cirrhosis-RD group than in the cirrhosis-NR group (all p<0.05). AST and TBil did not differ between the cirrhosis subgroups (all p>0.05) but were significantly higher in the subgroups than in the NCs (all p<0.01).

Figure 1 shows representative T1 and RBF maps of the two subgroups and the NCs. Table 3 presents the MRI parameters of each group. The renal cortical T1 (cT1) values, medullary T1 (mT1) values and cortical RBF values showed no significant differences between the left and right kidneys (all P > 0.05; Supplementary Table 1 ). The average measurements of both sides of each participant’s kidney were used for the final analysis.

The renal cortical T1 (cT1) values, medullary T1 (mT1) values and cortical RBF differed significantly among the three groups (all p<0.001). Subsequent post hoc tests confirmed significant differences between the NCs and cirrhosis-NR group for renal cT1 and mT1 values (all p<0.001), between the NCs and cirrhosis-RD group for renal mT1 values (p<0.01), and between both cirrhosis subgroups for renal cT1 and RBF values (all p<0.05; Figure 2A ). Cortical RBF values were weakly to moderately correlated with Scr (r= −0.334, p=0.009) and eGFR (r=0.483, p=0.003; Figure 2B ).

Table 4 and Figure 3 show the ROC curves of the T1 and RBF values for predicting renal impairment in patients with liver cirrhosis. Both renal cT1 and mT1 had high diagnostic performances in differentiating the cirrhosis-NR group and NCs (AUCs: 0.883 and 0.826, respectively; p=0.3362). Renal RBF performed better than did mT1 in differentiating between the cirrhosis-RD group and NCs (AUCs: 0.978 and 0.843, respectively; p=0.0076) and better than did cT1 in differentiating between the cirrhosis-RD and cirrhosis-NR groups (AUCs: 0.978 and 0.744, respectively; p=0.0187).

The classification tree obtained via multivariate CART analysis included renal T1 and RBF values and suggested the best sequence of their application in a diagnostic algorithm. After including renal cT1, mT1, and cortical RBF, only renal cT1 and RBF were selected to create the classification tree ( Figure 4 ). In the first step, the cortical RBF decrease (≤ 204 mL/100 g/min) was used to classify nine patients with cirrhosis (90%) as potentially having RD. For the remaining 51 participants who had no decrease in RBF (> 204 mL/100 g/min), renal cT1 was evaluated in the second step. Twenty-five participants exhibited decreased renal cT1 values (cortical T1 ≤ 1184 ms), indicating cirrhosis-NR, and 17 (73.9%) exhibited high renal cT1 (> 1184 ms), indicating they were the NCs. Thus, the diagnostic algorithm correctly identified cirrhosis-RD in 9 of 10 patients and cirrhosis-NR in 25 of 27 patients, yielding 85% accuracy. Table 5 shows the diagnostic performances for detecting NCs, cirrhosis-NR and cirrhosis-RD. Renal mT1 and cortical RBF were manually selected to create the classification tree with 80% accuracy ( Supplementary Figure 1 ).

The ICCs of the cT1, mT1 and cortical RBF values were 0.936 (95% confidence interval [CI], 0.894–0.962), 0.888 (95% CI, 0.812–0.933), and 0.947 (95% CI, 0.912–0.968), respectively. Supplementary Figure 2 shows the Bland-Altman plots.