Our institutional ethics review board approved this retrospective, single-center study. All patients gave written, informed consent for MRI and interventional procedures. Only patients with confirmed PSC according to EASL guidelines [11] who underwent GA-MRI between October 2007 and March 2022 were included. Patients with small-duct PSC, secondary sclerosing cholangitis, confounding liver illnesses (autoimmune hepatitis (AIH), Primary biliary cholangitis (PBC), Alcoholic liver disease (ALD), etc.), current or prior malignancy, who underwent orthotopic liver transplantation (OLT) prior to MRI, were under the age of 18, and/or had incomplete GA-MRI exams were excluded. Only patients with PSC-associated malignancies, i.e., cholangiocarcinoma, hepatocellular carcinoma (HCC), and gallbladder cancer were enrolled in the study. Patients with current or prior malignancies unrelated to PSC e.g., lung cancer, etc., were also excluded. Thus, our final cohort was comprised of adults with large-duct PSC who had at least one multiparametric GA-MRI, including conventional 2D- and 3D-T2-weighted-MRCP (Fig. 1 Flowchart).

Flowchart Between 2007 and 2022, 8564 patients underwent a standardized 3.0 Tesla contrast-enhanced MRI of the liver. Of these, 1312 patients were excluded because they were imaged using a contrast agent other than gadoxetic acid. A diagnosis other than sclerosing cholangitis further eliminated 7016 patients. Among patients with sclerosing cholangitis, additional exclusion occurred due to: secondary sclerosing cholangitis (SSC) (70), either overlap syndrome and/or small-duct PSC (14), incomplete HBP (10), previous OLT (11), age under 18 years (3), and prior malignancy (5)

Demographic and clinical data obtained from electronic medical records included patient age, gender, body mass index (BMI), date of, and indication for MRI, follow-up imaging exams, duration of PSC, and the presence of liver cirrhosis or concomitant inflammatory bowel disease (Table 1). Laboratory tests performed within two weeks of MRI, plus clinical scores that indicated disease severity, including MELD [12], Revised Mayo-Risk-Score (RMRS) [4], Fib-4 [13], APRI [14], ALBI [15], UK-PSC risk scores [16] and Prognostic Index of the Amsterdam-Oxford model (PI-AOM) [17], were recorded (Table 2). Clinical events that occurred after inclusion, orthotopic liver transplantation (OLT), death, cause of death, and cirrhosis decompensation defined as the occurrence of variceal bleeding, ascites, hepatic encephalopathy, or hepatorenal syndrome were recorded.

Patients entered the survival analyses at the time of GA-MRI. In March 2022, patient records were censored at date last seen, if they did not experience any sequelae. Survival status (alive, deceased), and date and type of whichever liver-related event first occurred were recorded (Table 1). OLT, death and decompensation (including encephalopathy, ascites, bleeding of oesophageal varices) were recorded as sequelae. The new occurrence of cholangiocarcinoma (CCA), gallbladder cancer, or hepatocellular carcinoma (HCC) were noted but not considered as sequelae if patients were still alive at the end of the study. Diagnostic or therapeutic endoscopic retrograde cholangiopancraticography (ERCP) was not considered as an event.

PSC severity and expected prognosis were based upon previously validated scores, including the RMRS, Fib-4, APRI, ALBI, Short-Term UK-PSC risk score, and PI-AOM for PSC [7, 16, 18, 19] (Table 2). We further classified these continuous scores into categorical groups, e.g., RMRS: low-risk (≤ 0); intermediate-risk (> 0 and < 2); and high-risk (≥ 2) groups [20, 21]. Binary classification of the Fib-4 was 0 (≤ 1.3) and 1 (> 1.3) [22], and the APRI was 0 (≤ 1.17) and 1 (> 1.17) [22]. The ALBI score was grouped into ≤ −2.60 (Grade 1), >  −2.60 to ≤ −1.39 (Grade 2), and > −1.39 (Grade 3) [23, 24], and the PI-AOM into low-risk, low – intermediate-risk, moderate-risk, and high-risk [25] (Table 2).

All examinations were performed on a 3 Tesla MR (MAGNETOM Trio Tim, or PrismaFit, Siemens). T2-weighted-MRCP was performed according to Hoeffel et al’s protocol and adhered to International PSC Study Group recommendations [26, 27]. MRCP images included a respiratory-triggered, 3D, heavily T2-weighted sequence in the coronal plane and a breath-hold, thick slab, single-shot, 2D, heavily T2-weighted sequence in the coronal and oblique coronal projections. Dynamic MR images were obtained in the, following transverse plane covering the whole liver during end-expiratory breath hold. T1-weighted VIBE images were obtained pre-contrast and during arterial (using automatic bolus tracking, using TWIST sequence (TR: 68.77, TE: 1.52, FA: 30°, slice thickness: 20 mm, FOV: 350, matrix 148 × 192, and only one NEX), portal venous (70 s), transitional (300 s) and hepatobiliary (20 min) phases (HBP) after injection of gadoxetic acid (Primovist® in Europe, Eovist® in USA). In addition, an axial T2-weighted sequence, an axial T2- weighted sequence with fat-suppression, axial in- and out-of-phase T1-weighted gradient echo sequence and diffusion-weighted sequences with three b-values (50, 300 and 600) and ADC map, and axial and coronal plane HBP images (with a flip angle of 20° and 35°) were obtained. The examination parameters for the whole MRI exam are given in Table 1S.

MRI exams were anonymized, then independently evaluated on a commercially available PACS workstation by five readers, R-A, R-B, R-C, R-D and R-E with 2, 3, 4, 6 and > 20 years of experience in abdominal radiology, respectively. To assess the intra-reader agreement, R-C and R-E reviewed the images twice, at least 12 weeks apart.

Blinded to all clinical data except PSC diagnosis, the radiologists assessed the ANALI scores for each patient. The individual parameter for Anali scores, including intrahepatic bile duct dilatation (IHBD) was assigned as = 0 if any duct was 3 mm or smaller, = 1 if any duct was 4 mm, and = 2 if ducts was 5 mm or larger. Liver dysmorphia was considered present, score = 1, if there was atrophy, lobulation of the liver contour and/or increased caudate-to-right liver lobe ratio [6, 7]. Otherwise, dysmorphia was considered absent, i.e., score = 0. Portal hypertension (PH) was considered present if there were collateral vessels, with or without splenomegaly, score = 1, were observed. Otherwise, PH was scored = 0, indicating its absence. Liver enhancement was assessed on both the arterial (AP)- and hepatobiliary-phase images (HBP). If parenchymal enhancement was uniform, it was scored = 0. Otherwise, heterogeneous liver parenchymal enhancement was scored = 1.

Then Anali scores were calculated as follows: ANALINoGd = (1 × dilatation of intrahepatic bile duct (IHBD)) + (2 × liver deformity) + (1 × PH), range 0 to 5, and ANALIGd = (1 × liver deformity) + (1 × parenchymal enhancement heterogeneity), range 0 to 2 [6, 7].

Thereafter, Anali scores were dichotimized as follows, ANALINoGd: low risk (0–2 points) and high risk (3–5 points), i.e., binary, ≤ 2 and > 2. ANALIGd in AP and HBP: low risk and high risk, i.e., binary ≤ 1 and > 1.

On 20-minute HBP images, patients were also dichotomized into normal contrast excretion through the biliary system at 20 min (NFS) or impaired excretion (PFS), i.e., no contrast seen either to: first-order left hepatic duct (LHD) or right hepatic duct (RHD) or common hepatic duct (CHD/hilum) or common bile duct (CBD) or none at all at 20 minutes [10]. For further details regarding PFS, please refer to Poetter-Lang et al [10].

Furthermore, estimated splenic volume was calculated as [mL] = 30 + 0.58 × L × D × T [28]. A cut-off value of 381.1 cm³ was chosen to differentiate normal-sized from enlarged spleens (Table 2) [29].

Metric data are presented as means ± standard deviations or median and quartile, depending upon their distribution. Nominal data are presented as absolute frequencies and percentages. Categorical data were evaluated by the chi-squared-test or the Fisher’s exact test.

Inter- and intrareader agreement between radiologists were assessed using Fleiss,’ and Cohen’s kappa, respectively. The Fleiss’ kappa was determined separately for each parameter, as well as low-risk vs high-risk (i.e., binary) for all 3 ANALI scores, indicating inter-reader agreement. For reader E, the intra-reader Cohen’s kappa was obtained. Then 95% confidence intervals (CI) were calculated for each value.

The kappa values < 0 indicated poor, 0.00–0.20 slight, 0.21–0.40 fair, 0.41–0.60 moderate, 0.61–0.80 substantial, and 0.81–1.00 almost perfect agreement [30]. Event-free survival was defined as the time interval from MR diagnosis to the first liver-related event occurrence. Kaplan-Meier estimates were performed, survival curves compared with log-rank test, when applicable. Univariate Cox proportional hazards regression analysis was performed to evaluate the association between outcomes and binary data for all 4 imaging parameters, i.e., PFS vs NFS, and (low-risk vs high-risk) ANALI scores, clinical scores and laboratory parameters. Multivariate analysis with adjustment for age and sex was also performed.

For event-free survival analysis, we calculated the results of each reader separately and then averaged them, yielding a mean event-free survival for the PFS and each ANALI score.

For survival analyses, cohort data were dichotomized into low- and high-risk scores, for ANALINoGd, i.e., ≤ 2 and > 2, and for both the ANALIGdAP and ANALIGdHBP, i.e., ≤ 1 and > 1. The level for statistical significance was set at p < 0.05. Statistical analyses were performed using R Studio (Version 1.4.1717) and IBM SPSS (version 26).