Multiparametric MR Index for the Diagnosis of Non-Alcoholic Steatohepatitis in Patients with Non-Alcoholic Fatty Liver Disease

Non-alcoholic steatohepatitis (NASH) is a complex disease consisting of various components including steatosis, lobular inflammation, and ballooning degeneration, with or without fibrosis. Therefore, it is difficult to diagnose NASH with only one imaging modality. This study was aimed to evaluate the feasibility of magnetic resonance imaging (MRI) for predicting NASH and to develop a non-invasive multiparametric MR index for the detection of NASH in non-alcoholic fatty liver disease (NAFLD) patients. This prospective study included 47 NAFLD patients who were scheduled to undergo or underwent ultrasound-guided liver biopsy within 2 months. Biopsy specimens were graded as NASH or non-NASH. All patients underwent non-enhanced MRI including MR spectroscopy (MRS), MR elastography (MRE), and T1 mapping. Diagnostic performances of MRS, MRE, and T1 mapping for grading steatosis, activity, and fibrosis were evaluated. A multiparametric MR index combining fat fraction (FF), liver stiffness (LS) value, and T1 relaxation time was developed using linear regression analysis. Receiver operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic performance of the newly devised MR index. Twenty NASH patients and 27 non-NASH patients were included. Using MRS, MRE, and T1 mapping, the mean areas under the curve (AUCs) for grading steatosis, fibrosis, and activity were 0.870, 0.951, and 0.664, respectively. The multiparametric MR index was determined as 0.037 × FF (%) + 1.4 × LS value (kPa) + 0.004 × T1 relaxation time (msec) −3.819. ROC curve analysis of the MR index revealed an AUC of 0.883. The cut-off value of 6 had a sensitivity of 80.0% and specificity of 85.2%. The multiparametric MR index combining FF, LS value, and T1 relaxation time showed high diagnostic performance for detecting NASH in NAFLD patients.

MR parameters and histopathologic components. Diagnostic performance of MRS for grading steatosis is summarized in Table 2. Using MRS, the mean area under the curve (AUC) for grading steatosis was 0.870. Diagnostic performances of MRE and T1 mapping for grading activity (ballooning and lobular inflammation) and fibrosis are summarized in Table 2. Using MRE, the mean AUCs for grading activity (ballooning and lobular inflammation) and fibrosis were 0.825 and 0.951, respectively. Using T1 mapping, the mean AUCs for grading activity and fibrosis were 0.664 and 0.615, respectively.
In subgroup analysis, the mean T1 value was significantly higher in group B (mean FF MRS > 15%) than in group A (mean FF MRS ≤ 15%) for ≥ F2 (significant fibrosis) patients (1133.9 ± 97.0 vs. 935.9 ± 79.6 msec, p = 0.001). AUCs of T1 mapping for grading ≥ F2 (significant fibrosis) were 0.837 and 0.714 in group A and B, respectively. These results indicated that hepatic fat is a confounding factor of T1 value estimation.

MR index. A multiparametric MR index combining fat fraction (FF) (measured on MRS), liver stiffness (LS)
value (on MRE) and T1 relaxation time (on T1 mapping) was determined as 0.037 × FF (%) + 1.4 × LS value (kPa) + 0.004 × T1 relaxation time (msec) −3.819 (Fig. 1). Receiver operating characteristic (ROC) curve analysis revealed an AUC of 0.883 and an optimal cut-off value of 4.6 with a corresponding sensitivity of 95.0% and specificity of 77.8% (Table 3). The specificity was relatively low whereas the sensitivity was high. To exclude patients who do not need a liver biopsy, because specificity is more important than the sensitivity, a cut-off value of 6 exhibiting a relatively high specificity while maintaining the sensitivity was used (sensitivity of 80.0% and specificity of 85.2%) ( Table 3, Fig. 2).
In subgroup analysis, the MR index showed higher diagnostic performances in both group A and B than in the entire group. In group A (mean FF MRS ≤ 15%), the MR index showed an AUC of 0.909 and an optimal cut-off value of 4.3 with a corresponding sensitivity of 100% and specificity of 83.3%. In group B (mean FF MRS > 15%), the MR index showed an AUC of 0.901 and an optimal cut-off value of 4.6 with a corresponding sensitivity of 100.0% and specificity of 66.7% (Table 3).    www.nature.com/scientificreports www.nature.com/scientificreports/ or without fibrosis. Therefore, we devised a new multiparametric MR index combining FF, LS value, and T1 relaxation time. Our newly devised MR index predicted NASH in NAFLD patients with a sensitivity of 80.0% and a specificity of 85.2%.
NAFLD is a term used to describe fat accumulation in the liver with no history of excess alcohol intake or other liver disease. NASH is a subgroup of NAFLD, characterized by hepatocellular injury, inflammation, and ballooning with or without fibrosis. There are several staging and grading systems for NASH. The most widely used system is the NASH Activity Score (NAS), which was originally developed as a tool for evaluating therapeutic outcomes in NAFLD patient trials. It has proven useful for comparative or interventional studies. However, it is less beneficial as a diagnostic tool for NASH and has low prognostic value because it does not include fibrosis 16 . The more recently developed SAF scoring system may be more accurate for identifying NASH and be more relevant for long-term prognostication 17,18 .
Hepatic fat is an essential component for diagnosing NAFLD/NASH. There have been several efforts to identify hepatic steatosis using various imaging techniques including ultrasonography 19 , unenhanced computed tomography 20,21 , MRI 22 , and MRS 23,24 . MRS is considered the most precise non-invasive imaging technique for hepatic fat quantification. In the past, MRS was performed with free-breathing, resulting in long acquisition times and erroneous data acquisition due to motion artifacts 25 . Recently, high-speed T2-corrected multi-echo (HISTO) MRS was introduced for T2-corrected hepatic fat measurement with a single breath hold 26,27 . In our study, MRS showed high mean AUC of 0.870 for grading steatosis in NAFLD patients.  Table 3. Diagnostic performance of multiparametric MR index.

Figure 2.
An example of using the MR index to predict NASH. A 51-year-old man with clinically suspected NASH who underwent percutaneous liver biopsy and MR imaging. Fat fraction measured on MR spectroscopy was 11.3%, liver stiffness value measured on MR elastography was 3.13 kPa, and T1 relaxation time measured on T1 mapping was 877.2 msec. The non-invasive multiparametric MR index predicted that the patient was a non-NASH patient. Histopathologic analyses of the patient's biopsy specimens revealed grade 1 steatosis, grade 0 balloon degeneration, grade 1 lobular inflammation, and grade 1 fibrosis. Therefore, he was classified as a non-NASH patient according to the SAF scoring system.
NAFLD ranges from simple steatosis to NASH, which can progress to cirrhosis and HCC. Although liver biopsy is considered the only reliable method for distinguishing NASH from simple steatosis, there is the potential risk of sampling error and complications such as pain, bleeding, and although extremely rarely, even death. These drawbacks of liver biopsy have increased the demand for alternative and noninvasive methods for diagnosis of NASH. These include non-invasive models composed of demographic and laboratory parameters, such as aspartate aminotransferase to platelet ratio index (APRI), fibrosis-4 index (FIB-4), BARD score, and NAFLD fibrosis score (NFS) 18 . Non-invasive imaging techniques such as acoustic radiation force impulse (ARFI), transient elastography (FibroScan), shear wave elastography (SWE), and MRE have also been studied to detect fibrosis in NAFLD patients 18 . Among them, MRE showed the highest diagnostic accuracy for the grading and diagnosis of fibrosis in NAFLD patients in meta-analysis and prospective studies 28,29 . In our study, MRE showed high mean AUC of 0.951 for grading fibrosis in NAFLD patients.
T1 mapping was originally used for non-invasive evaluation of myocardial fibrosis, which is characterized by edema (increase in tissue water) and increased extracellular matrix (ECM) remodeling (increase in collagen volume fraction) after an acute ischemic event 30,31 . Several studies have assessed this technique in chronic liver disease patients 11,32,33 . In liver fibrosis, increased edema and ECM remodeling occur after hepatocyte injury, resulting in lengthened T1 relaxation time. The T1 mapping technique directly measures T1 relaxation time in milliseconds. Banerjee et al. demonstrated that T1 relaxation time was strongly correlated with liver fibrosis in 79 patients who underwent liver biopsy 11 . T1 mapping had an AUC of 0.94 with a sensitivity of 86% and a specificity of 93% for differentiating healthy people and patients with no fibrosis from those with fibrosis. In our study, compared with MRE, T1 mapping showed relatively low diagnostic performances for grading fibrosis (mean AUC, 0.615 vs. 0.951) and activity (mean AUC, 0.664 vs. 0.825). This result may be due to the fact that T1 values are influenced by hepatic fat 34 . Therefore, subgroup analyses were performed to evaluate the diagnostic performance of T1 mapping for grading significant fibrosis (F2) and the diagnostic performance of MR index for detecting NASH. Diagnostic performance of T1 mapping improved in each subgroup (FF MRS ≤ 15% and > 15%) relative to the entire group. In particular, diagnostic performance was higher in group A (FF MRS ≤ 15%) than in group B (FF MRS > 15%). Diagnostic performance of the MR index also improved in each group relative to the entire group and was higher in group A than in group B. These results indicate that hepatic fat is a confounding factor of the T1 value, therefore the T1 value should be used with caution in patients with fatty liver of moderate or greater degree. Applying different MR indices according to hepatic fat component can overcome the limitation of T1 mapping. To the best of our knowledge, there is no new sequence yet to improve T1 mapping.
As noted above, MRI studies in NAFLD/NASH have focused primarily on the diagnosis of fat or fibrosis, which are components of NASH. Like non-invasive models using clinical data, if a non-invasive MR index is developed by combining MR parameters in NAFLD patients, it would provide an easily available prediction model in the clinical setting. Therefore, a new MR index was devised by combining FF, LS value, and T1 relaxation time measured by MRS, MRE, and T1 mapping, which are included in our MRI protocol for NAFLD patients. This new MR index predicted NASH as a SAF score with high diagnostic performance.
This study had several limitations. First, only a few patients limited to a single tertiary center using one MRI scanner were included. Although a small cross-vendor validation study comparing the reproducibility of two difference MR scanners from Philips and GE Healthcare in 13 subjects demonstrated that LS value measurements were reproducible and had good consistency across two vendors 35 , a multicenter study with a larger number of patients using several different MRI scanners is needed to develop an MR index that can be widely used in various clinical settings. Second, our new devised MR index was not validated using a validation group due to the small number of patients. Another prospective study using a validation group of NAFLD patients is needed to validate the results of this study. Third, it is well known that hepatic iron deposition can be increased associated with liver disease 36 , and age, gender, and menopause for female may affect hepatic iron deposition 37,38 , which in turn may affect MRI measurements. However, it was not possible to perform subgroup analyses according to age, gender, and menopause due to small number of patients. We hope to study the effect of hepatic iron deposition on MRI measurements in the future with a larger number of patients. Fourth, in our study, a non-invasive index was developed using only MR parameters and linear regression analysis with three fixed MR parameters (FF, LS value, and T1 relaxation time). However, the aim of the study was to develop a simple MR index for predicting NASH in NAFLD patients. In the future, it is expected to devise a NASH index through a prospective study using multivariate regression analysis of clinical (demographic and laboratory) data and MR parameters.
In conclusion, MRI showed high diagnostic performance for detecting and grading the NASH components of steatosis, activity (lobular inflammation and ballooning degeneration), and fibrosis. Our newly devised MR index combining MRS, MRE, and T1 mapping also demonstrated high diagnostic performance for detecting NASH in NAFLD patients. Therefore, this multiparametric MR index may help diagnose NASH in NAFLD patients and potentially reduce the need for liver biopsy.

Materials and Methods
Study design. This clinical trial was a prospective study conducted in accordance with the Declaration of Helsinki at the Korea University Guro Hospital in Korea. The study was approved by the Institutional Review Board of the Korea University Guro Hospial (approval no.: KUGH16184) and written informed consent was obtained from each patient. This trial was registered into ClinicalTrials.gov on 22 November 2017. The study followed CONSORT guidelines in the reporting of results.

Patients.
Patients with clinically suspected NASH who were scheduled to undergo or underwent liver biopsy within 2 months were identified from October 2016 to June 2017. Patients 18 years and older who voluntarily participated in this study were included. Exclusion criteria were as follows: (a) other known causes of chronic liver disease such as chronic hepatitis B or C, autoimmune hepatitis, and primary sclerosing cholangitis; (b) use of (2020) 10:2671 | https://doi.org/10.1038/s41598-020-59601-3 www.nature.com/scientificreports www.nature.com/scientificreports/ steatogenic medications within the past six months; (c) significant alcohol consumption (more than 70 g per week for women and 140 g per week for men); (d) history of HCC; (e) pregnancy; and (f) contraindications to perform MRI. Finally, 47 patients were included in this study. Demographic and clinical data including age, gender, height, weight, and body mass index were obtained. Laboratory data within 1 month of the MR exam were also recorded. These are summarized in Table 1.
MR spectroscopy. Single-voxel MRS was performed using a prototypical HISTO MR spectroscopic technique. The HISTO technique is a modified stimulated echo acquisition mode (STEAM) sequence 27 . MRS spectra were obtained as the same method described in our previously published study 27 . A single voxel (20 × 20 × 20 mm) was placed in the right hepatic posterior segment by an experienced technologist (more than 10 years of experience in MRS) while avoiding large bile ducts, vessels, and focal hepatic lesions (Fig. 3a). Parameters included repetition time (TR), 3000 msec; mixing time, 10 msec; and 5 echo times (TEs) of 12, 24, 36, 48, and 72 msec. Each MRS acquisition was performed within 15 seconds during one breath hold. This process was repeated three times. Post-processing of the MRS data was performed using inline software of the MR vendor.
MR elastography. A pneumatic driver system (Resoundant, Rochester, MN, USA) was used to induce shear waves into the liver. A cylindrical passive driver was placed against the patient's right anterior chest wall at the level of the xiphoid process 8 . Continuous acoustic vibrations of 60 Hz were transmitted from the active driver to the passive driver through a flexible vinyl tube 8 . Measurement parameters of a prototypical MR elastography 2D spin-echo based echo-planar sequence were as follows: TR/TE, 100/47 msec; SPAIR fat suppression; field of view (FOV), 380 mm × 380 mm; matrix size, 100 × 100; slice thickness, 6 mm; and interslice gap, 1.2 mm. Four MR elastography sections were obtained in each patient and it took 11 seconds to obtain four slices. Patients were requested to hold their breath at the end of expiration. After the acquisition was completed, wave images were processed inline by the MR scanner to generate stiffness maps (elastograms) in kilopascals (kPa). Elastograms were displayed with a scale corresponding to the stiffness value.
T1 mapping. For T1 mapping, a prototypical 2-dimensional Look-Locker inversion recovery T1 mapping sequence based on fast low-angle shot (FLASH) with inline construction was performed. Three axial slices were obtained at the level of confluence of the hepatic veins to the inferior vena cava, portal hilum, and gallbladder www.nature.com/scientificreports www.nature.com/scientificreports/ fossa within a breath hold. For each slice the Look-Locker T1 mapping sequence applies a non-selective inversion recovery pulse followed by low-flip angle FLASH acquisitions of multiple inversion contrasts. Imaging parameters were as follows: TR/TE, 3 Measurement of liver stiffness on MRE. To measure liver stiffness (LS) values, two board-certified radiologists (C.H.L. and J.W.K. with 25 years and 6 years of experience in abdominal radiology, respectively), blinded to the clinical data and pathologic results, drew free-hand regions-of-interest (ROIs) along the liver margin by consensus, avoiding large vessels, liver edges, and motion artifacts. ROIs were manually drawn on wave images where wave propagations were regular and relatively free of reflections and interference patterns. ROIs drawn on wave images were copied and pasted onto the stiffness maps (Fig. 3b). One ROI per slice (total 4 slices) was measured, and mean LS values (in kilopascals) were used as representative values.
Measurement of T1 relaxation time on T1 mapping. To measure T1 relaxation time, the same two radiologists, blinded to the clinical data and pathologic results, drew free-hand ROIs along the liver margin on each slice of the T1 maps by consensus, avoiding major intrahepatic vessels and focal hepatic lesions (Fig. 3c). One ROI per slice (total 3 slices) was measured, and mean T1 relaxation times (in milliseconds) were used as representative values.
Statistical analysis. For clinical and laboratory data, the independent-sample t-test was used to compare continuous variables, and the chi-square (χ 2 ) test was used to compare categorical variables between the NASH and non-NASH groups.
To evaluate the diagnostic performances of MRS, MRE, and T1 mapping for grading each histopathologic component (steatosis, ballooning, locular inflammation, and fibrosis), ROC curve analyses were performed and AUCs were obtained.
A previous study demonstrated that the T1 value estimates may be affected by liver fat 34 . Therefore, subgroup analyses were performed to evaluate the diagnostic performances of T1 mapping and the MR index using ROC curve analyses. Patients were assigned to one of two groups based on mean FF MRS : group A (mean FF MRS ≤ 15%) and group B (mean FF MRS > 15%).

Development of the MR index.
A multiparametric MR index as a weighted linear combination of liver signal FF, LS value, and T1 relaxation time was developed using linear regression analysis to predict the SAF score (Fig. 1). ROC curve analysis was performed to evaluate the diagnostic performance of the MR index.