Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Jul 1.
Published in final edited form as: J Magn Reson Imaging. 2021 Feb 15;54(1):122–131. doi: 10.1002/jmri.27549

Automated Analysis of Multi-parametric MRI/MRE Exams for prediction of NASH

Bogdan Dzyubak 1, Jiahui Li 1, Jie Chen 1, Kristin C Mara 1, Terry M Therneau 1, Sudhakar K Venkatesh 1, Richard L Ehman 1, Alina M Allen 2, Meng Yin 1
PMCID: PMC8195849  NIHMSID: NIHMS1692252  PMID: 33586159

Abstract

Background:

Nonalcoholic fatty liver disease (NAFLD) affects 25% of the global population. The standard of diagnosis, biopsy, is invasive and affected by sampling error and inter-reader variability. We hypothesized that widely available rapid MRI techniques could be used to predict nonalcoholic steatohepatitis (NASH) non-invasively by measuring liver stiffness, with magnetic resonance elastography (MRE), and liver fat, with chemical shift-encoded (CSE) MRI. Besides, we validate an automated image analysis technique to maximize the utility of these methods.

Purpose:

To implement and test an automated system for analyzing CSE-MRI and MRE data coupled with model-based prediction of NASH.

Study Type:

Prospective.

Subjects:

83 patients with suspected NAFLD.

Field Strength/Sequence:

1.5 T using a flow-compensated motion-encoded gradient echo MRE sequence and a multi-echo CSE-MRI sequence.

Assessments:

The MRE and CSE-MRI data were analyzed by two readers (5+ and 1 years of experience) and an automated algorithm. A logistic regression model to predict pathology-diagnosed NASH was trained based on stiffness and proton density fat fraction, and the area under the receiver operating characteristic curve (AUROC) was calculated using 10-fold cross validation for models based on both automated and manual measurements. A separate model was trained to predict the NASH severity score (NAS).

Statistical Tests:

Pearson’s correlation, Bland-Altman, AUROC, C-statistic.

Results:

The agreement between automated measurements and the more experienced reader (R2 = 0.87 for stiffness and R2 = 0.99 for PDFF) was slightly better than the agreement between readers (R2 = 0.85 and 0.98). The model for predicting biopsy-diagnosed NASH had an AUROC of 0.87. The NAS-prediction model had a C-statistic of 0.85.

Data Conclusion:

We demonstrated a workflow that used a limited MRI acquisition protocol and fully automated analysis to predict NASH with high accuracy. These methods show promise to provide a reliable noninvasive alternative to biopsy for NASH-screening in populations with NAFLD.

Keywords: MRE, stiffness, PDFF, CSE-MRI NASH, liver

INTRODUCTION

Nonalcoholic fatty liver disease (NAFLD) is the most common chronic liver disease and affects around 25% of the world’s population.1 Its rise parallels the rise of obesity, and NAFLD is projected to become a leading cause of liver-related morbidity and mortality within 20 years.2 NAFLD is characterized by an increase in liver fat content (steatosis) and inflammation.3 Approximately 20% of NAFLD patients develop a more aggressive condition, nonalcoholic steatohepatitis (NASH).4 NASH may lead to cirrhosis and hepatocellular carcinoma.3 Timely intervention with lifestyle interventions or bariatric surgery in early stages of the disease can lead to regression of NASH and fibrosis.5–9 It is therefore important to diagnose patients promptly and monitor their disease course and treatment response. NAFLD is diagnosed by histologic assessment of steatosis, inflammation and hepatocyte ballooning.3 However, liver biopsy is invasive and has limited reproducibility due to the small sample size and subjective scoring by readers9,10 both of which limit its usefulness for evaluating the large population potentially affected by NASH. Instead, liver fibrosis can be accurately and reproducibly assessed with Magnetic Resonance Elastography (MRE) which measures liver stiffness10 and liver fat content can be assessed with chemical shift-encoded (CSE) MRI.11

Liver MRE is a well-established technique for the noninvasive staging of fibrosis.12–14 Multiple prospective studies and meta-analyses have demonstrated that MRE is the most accurate noninvasive predictor of fibrosis stage, when compared to biopsy, with an area under the receiver operating characteristic curve (AUROC) of >0.9 for all fibrosis stages.12,15,16 Ultrasound elastography techniques offer another noninvasive alternative. However, MRE has several advantages including a higher accuracy for detecting fibrosis in patients with NAFLD19–22 and an important ability to effectively image patients with high BMI17,18 which is common in NAFLD/NASH.3 MRE also has a high success rate,23 high reproducibility,13,24–26 and low operator dependence.18 It typically images approximately a quarter of the liver which leads to substantially reduced sampling error compared to biopsy which samples approximately 1/50,000 of the liver, and ultrasound elastography which samples 1/500 of the liver.27 MRE images are analyzed by trained readers who identify a liver tissue region of interest (ROI) which excludes vessels, tumors, and motion artifact, and subsequently analyze the wave images to exclude areas of interference, and the elastogram to exclude artefactual hot- and cold-spots.28 The manual ROI selection is the most substantial limitation to the technical repeatability of MRE.29 Also, it requires specialized reader training and 15–20 minutes of analyst time per exam. Chemical shift-encoded magnetic resonance imaging (trade names: IDEAL, mDixon, qDixon) works by acquiring images at several echo times (typically 6) and solving for the quantitative proton density fat fraction (PDFF) images which are used to characterize fat deposition in NAFLD.30 Additional image contrasts generated include water images, fat images and R2* images. PDFF is typically measured by placing ROIs in 1 to 4 central liver slices. A more involved technique of placing 9 ROIs in each anatomical liver segment can be used to improve repeatability at the cost of significantly increased analysis time.31 Hence, while a quantitative liver exam to assess liver stiffness and PDFF for staging of fibrosis and steatosis, consisting of MRE and CSE-MRI, can be acquired in less than 5 minutes, analysis time is relatively long and limits adoption.

Thus the aim of this work was to develop a fast (<5 minute) reproducible automated method to calculate liver stiffness and PDFF from the combined MRE and CSE-MRI exam. We also aimed to use these two quantitative parameters to fit a model to predict biopsy-diagnosed NASH and NAFLD activity score (NAS). Our hypothesis was that the combination of liver stiffness and proton density fat fraction (PDFF), measured using a reproducible automated method, could be used to noninvasively predict the presence of NASH with high accuracy.

METHODS

Subjects

Under an IRB-approved prospective study, adult patients with obesity who were evaluated for bariatric surgery were enrolled between 2015 and 2018. Following informed consent, MRE and CSE-MRI exams were performed prior to surgery. Liver biopsy was obtained intraoperatively for the histological assessment of NAFLD. The exclusion criteria consisted of decompensated liver cirrhosis, history of liver transplantation, and excessive alcohol consumption within 6 months of study.

MRI Acquisition

The image acquisitions were performed on 1.5T whole-body scanners (GE Healthcare, Milwaukee, WI) with patients imaged in the supine position. For the MRE exam, the passive driver was placed against the anterior body wall over the right lobe of the liver and held in place with an elastic band wrapped around the body. Continuous acoustic pressure waves were generated at 60 Hz by an active driver outside the scanner room and delivered to the passive driver through a 7.6-m-long plastic tube. The acquisition was performed using a flow-compensated gradient-echo MRE sequence; axial imaging plane; 44–50-cm field of view; 256 × 64 acquisition matrix; 30° flip angle; 4 10-mm contiguous sections prescribed through the widest portion of the liver imaged sequentially; repetition time/echo time = 50/20.2 ms; 16.7-ms through-plane motion sensitizing gradient (MSG); MSG sensitivity = 10.1 μm per radian; four time offsets; bandwidth = ±31.25 kHz. The imaging time was 54 seconds split into four breath holds performed at the end of expiration. Patients fasted for at least 4 hours before the exam. The images yielded by an MRE acquisition, including the anatomical magnitude image, the wave image derived from the motion-encoded phase, and the quantitative elastogram, are shown in Figure 1.

Figure 1:

Figure 1:

Open in a new tab

Magnetic Resonance Elastography (MRE) images. A) Magnitude image showing anatomy. B) Wave image showing encoded motion. C) Quantitative elastogram with stiffness values at every voxel.

CSE-MRI was performed using a fast gradient-echo sequence acquisition with the following parameters: 6 echoes evenly spaced between 1.1 ms and 6.38 ms; single axial imaging plane; 35–50-cm field of view; 256 × 160 acquisition matrix; 7° flip angle; slices prescribed to cover the majority of the liver (64 slices, thickness 8 to12.6-mm; TR/TE1/TE2 = 100/2.13/4.94 ms; receiver bandwidth = ±83.33 kHz. The imaging time was 22 seconds split into two breath holds. The image contrasts were inverted on the scanner using the IDEAL-IQ (GE Healthcare, Milwaukee, WI) algorithm to calculate quantitative images of the water, fat, and fat fraction (PDFF) shown in Figure 2.

Figure 2:

Figure 2:

Open in a new tab

CSE-MRI images reconstructed from 6 acquired echoes. A) Fat-only image. B) Water-only image. C) Proton density fat fraction (PDFF) image. The slice has been selected by the automated method to match the slice in Figure 1.

Histology

Biopsy was obtained intraoperatively during the weight reduction surgery from the right liver lobe within 1 month of the MRI examination. One to four cores were extracted with a combined length of approximately 1.5 – 5.5 cm. All liver biopsy specimens were reviewed by our institution’s liver pathologists as part of standard clinical care. In addition, a study pathologist with NASH expertise, blinded to the clinical and imaging results, provided a second interpretation. In case of discrepancy, a third pathologist was invited for review. The disease severity was estimated using the NAS.32

Manual analysis

For MRE, the ROIs were drawn according to Quantitative Imaging Biomarker Alliance (QIBA) MRE protocol.33 A single region of interest (ROI) was drawn in each slice to encompass as much liver area as possible while excluding regions of 1) non-liver tissue, 2) low wave amplitude (assessed visually), 3) within 2 pixels of the liver’s edge, and 4) masked out by the multi-modal direct inversion (MMDI) confidence map thresholded at 0.95. The weighted mean of the elastogram voxels within the ROI across all slices was then reported as the liver stiffness. MRE analysis was performed by an expert reader with over 5 years of experience (**) and an experienced reader with 1 year of experience (**).

For the PDFF measurement, two readers drew circular ROIs in each of the 9 anatomical liver segments. The fat fractions calculated in the segments were averaged into a liver PDFF measurement.31 PDFF was measured by two expert readers with over 5 years of experience (** and **).

Automated Analysis

Liver stiffness and PDFF were also calculated using an automated method. For MRE, the approach has been described previously.34 Briefly, in all 4 MRE slices, the liver is segmented based on position (interior to body fat, and on the right side of the body) and intensity (darker than body fat but brighter than background). Then, localized areas of very high or low stiffness are removed as potential artifacts. In this work, we extended the method to analyze PDFF images. The method selected a subset of slices for analysis by calculating normalized mutual information between the 4 MRE slices and a sliding window of 4 PDFF slices. Mutual information is a cross-modality registration metric which assesses the correlations between voxel intensities. It has the advantage of not relying on tissue intensities having the same relative intensities across modalities.35 A stack of 4 fat images with the highest cumulative mutual information with the MRE volume was selected.

The preliminary segmentation was performed in fat images, which have a high contrast between the liver and the body wall (Figure 3). Then an outlier removal step was applied to exclude blood vessels and cavities from the ROI. In this step, 20% of the voxels with the highest absolute differences from the liver mean intensity were removed from the mask, and the mask was morphologically closed with a 5-pixel element. The morphological closing acts as a spatial constraint, which ensures that only large contiguous structures (greater than approximately 30 pixels) are removed, while individual “noisy” voxels are added back to the mask. The outlier removal step was performed in parallel based on the water image (to remove body cavities) and the PDFF image (to remove blood vessels). The weighted mean PDFF was then calculated for the liver. The automated pipelines for MRE and PDFF are shown in Figure 4.

Figure 3:

Figure 3:

Open in a new tab

Automated ROI selection procedure for CSE-MRI. A) Fat image with high contrast between the liver and the body wall. B) Probability function for a voxel belonging to the background (red), the abdominal wall (blue), the liver (green), or none of the above based on spatial location. C) Probability of a voxel belonging to the same classes based on intensity. D) Final ROI calculated based on combined probabilities and outlier cleanup.

Figure 4:

Figure 4:

Open in a new tab

Flowchart of automated processing of MRE and CSE-MRI. The MRE analysis uses an additional seed-driven Random Walker segmentation to refine the contour, as well as exclusion of localized soft/stiff regions based on the elastogram and wave images. CSE-MRI analysis has additional slice-selection and outlier exclusion steps.

NASH Prediction Model

A machine learning model based on logistic regression was constructed to predict the probability of biopsy-diagnosed NASH, using two parameters - liver stiffness and PDFF. A secondary model was fit to predict the NAFLD activity score (NAS), a metric used to describe NAFLD severity. NAS ranges from 0 to 8 and is the sum of histologic grades of steatosis, inflammation and ballooning. The performance of models trained independently on measurements from reader 1, reader 2, and the automated algorithm was compared.

Statistical Evaluation

The agreement in liver stiffness and PDFF measurements was evaluated using the coefficient of determination (R2) and Bland-Altman analysis. Measurements were compared between the two readers, and the automated method against each reader. The performance of NASH and NAS score prediction models was evaluated using AUROC and the C-statistic, respectively, in a ten-fold cross validation setup. Statistical analyses were performed in SAS version 9.4 (SAS Institute, Cary, NC) and R statistical software version 3.2.0 (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

A total of 83 subjects (mean (SD) age: 47 (±11), BMI: 47 (±9), 83% female) from our cohort had successful biopsy and MRE and CSE-MRI exams analyzed by two readers. The mean (SD) shear stiffness and fat fraction measured by Reader 1 (more experienced) were 2.6 (0.7) kPa and 13 (7)%, respectively. Automated ROIs for MRE and CSE-MRI were reviewed for every exam and all ROIs were found to be satisfactory. Thus, no manual modifications were made. The automated ROI tool successfully analyzed all of these exams. Based on biopsy, 34 patients had NASH, 25 had simple steatosis and 24 had normal liver. The disease activity (NAS) among those with NASH was as follows: 2 (n=3), 3 (n=14), 4 (n=11), 5 (n=4) and 6 (n=2).

The stiffness and PDFF measurements based on automated ROIs had a higher agreement with the expert reader (R2 = 0.87 for stiffness and R2 = 0.99 for PDFF) than the expert and experienced readers had with each other (R2 = 0.85 for stiffness and R2 = 0.98 for PDFF). The correlation and Bland-Altman plots for stiffness agreement are shown in Figures 5 and 6, and for PDFF in Figures 7 and 8.

Figure 5:

Figure 5:

Open in a new tab

MRE magnitude (a) and elastogram (b) images, and CSE-MRI PDFF (c) and water (d) contrasts with ROIs overlaid. On MRE, the ROI avoids major vessels, areas of low acoustic wave amplitude (checherkboard in b) and stiffness outliers (red areas in b). On CSE-MRI, the ROI avoids major vessels and any other outliers detected on either contrast images. The amount of tissue sampled by the ROI was similar since CSE-MRI had half the slice thickness of MRE, while MRE had only partial penetration into the liver due to acoustic wave attenuation.

Figure 6:

Figure 6:

Open in a new tab

Stiffness correlation between the algorithm and each reader as well as between the two experienced readers.

Figure 7:

Figure 7:

Open in a new tab

Bland-Altman of stiffness differences between the algorithm and each reader, as well as between the two experienced readers.

Figure 8:

Figure 8:

Open in a new tab

PDFF correlation between the algorithm and each reader as well as between the two experienced readers.

The model predicting biopsy-diagnosed NASH had an AUROC of 0.87. The Chi-Squared was 13.13 (p=0.0003) for liver stiffness and 54.77 (p<0.0001) for PDFF, indicating that PDFF had a higher weight in the model. The performance was the same to two digits for models based on the manual and the automated measurements. The ROC curves are shown in Figure 9. The NAS prediction had a cross-validated C-statistic of 0.85 for both manual and automated measurements.

Figure 9:

Figure 9:

Open in a new tab

Bland-Altman of PDFF differences between the algorithm and each reader, as well as between the two experienced readers.

DISCUSSION

In this work, we used a limited 5-minute protocol to acquire MRE and CSE-MRI image data, and demonstrated that quantitative liver stiffness and fat fraction calculated from them can be used by a machine learning model to predict NASH with high accuracy. Additionally, we validated an automated ROI selection tool that can be used in combination with the NASH prediction model to create a fully-automated workflow with excellent agreement to trained readers.

The analysis of MRE images is a multi-step process that requires special training.28 It typically takes 10–20 minutes to analyze a 4-section MRE exam and requires substantial experience to be considered expert. The MRE analysis algorithm part of the automated toolkit has been described previously.34 It has been used in the clinical workflow to minimize inter-reader variability and reduce the workflow process to 5 minutes, which includes manual verification. While the manual measurement of PDFF takes less training and time than MRE analysis. Automation of the full pipeline has special merit as it allows the NASH prediction to be provided as part of the workflow without querying the clinical record.

The new automated ROI selection tool for PDFF, similarly to the MRE analysis tool, relies on knowledge of relative spatial positions and intensities of major structures like the body wall, liver, and kidneys. CSE-MRI images do not contain motion artifact often present in MRE and have high contrast between the body wall and the liver. However, the contrast between the liver and other internal organs (kidneys, intestines, lungs, heart) can be close to zero which makes traditional segmentation methods fail. Spatial information is used to improve the segmentation and is an effective solution in slices prescribed in the wide part of the liver. In slices which are placed too superior, near the lungs and heart, or too inferior, with little liver area, segmentation remains extremely challenging. We found that this issue was successfully resolved by selecting only a subset of 4 slices for the PDFF measurement in the widest part of the liver, which is similar to the manual workflow. In this study, we selected slices by anatomically registering to the 4 MRE slices that are prescribed in the wide part of the liver. The acquisition duration for this workflow can be further expedited by acquiring only the desired slices Overall, the agreement in stiffness and PDFF measurement was excellent between the algorithm and human readers. The discrepancies increased for higher stiffness and PDFF values both for algorithm vs reader and reader vs reader because blood vessels, which may not be fully excluded from the ROIs, have apparent stiffness/PDFF values similar to healthy liver. Also, the liver becomes more heterogeneous at higher stiffness values leading to lower inter-reader agreement..

PDFF is known to reliably characterize steatosis (fat deposition in the liver). However, PDFF does not allow simple steatosis to be distinguished from NASH, which involves inflammation and tissue damage.36 In patients with known steatosis, MRE can be used to predict NASH with high accuracy.37 In this study we used a combination of liver stiffness and PDFF to also predict NASH with high accuracy. A previous study used 3D multifrequency MRE to calculate stiffness and damping ratio, and combined them with PDFF to predict NASH obtaining an AUROC of 0.85.38 The current study demonstrates that screening could also be performed using standard MRE which is available at over 1500 sites worldwide. The minimal apparent improvement in AUROC with respect to the 3D MRE study (0.87 vs 0.85) is most likely inconsequential and could be attributable to us using 10-fold cross validation, while the study using 3D MRE treated the same group of patients as a dedicated test cohort. Our secondary model predicted the NAS score and may facilitate the monitoring of patients’ disease progression and treatment response in clinical trials. Similarly, the raw predicted probability of NASH may also be useful for this purpose.

Limitations

Our study population consisted exclusively of high BMI patients who had undergone bariatric surgery, predominantly women (81%). The subjects typically had a large amount of abdominal fat and varying degrees of hepatomegaly which may affect segmentation performance. BMI is not known to correlate with liver stiffness or PDFF, and was high for both NASH and non-NASH subjects. Due to indications for bariatric surgery, patients included in this study mostly had mild to moderate liver disease. The dataset will benefit from additional patients with cirrhosis (liver stiffness >5kPa) and iron overload. However, since patients with advanced disease are easier to classify, we do not believe that their sparsely adversely affects the model to a substantial degree.

This study used exclusively GE scanners due to availability at our institution. However, both the MRE and PDFF acquisitions are available from all major vendors, and our automated method has been successfully applied to them in currently unpublished work.

Conclusion

This study has demonstrated non-invasive high-accuracy prediction of biopsy-diagnosed NASH based on automated analysis of a 5 minute quantitative imaging protocol. The automated liver stiffness and PDFF measurements had excellent agreement with an expert human reader, do not suffer from reader variability, and provide a clinical workflow access point for the NASH prediction model.

Figure 10:

Figure 10:

Open in a new tab

ROC curves for prediction of biopsy-diagnosed NASH based on stiffness and PDFF measured by the algorithm and each experienced reader.

Grant support:

EB017197, AA026887, W81XWH-19-1-0583-01, K23DK115594, R37 EB001981

REFERENCES

  • 1.Younossi Z, Koenig A, Abdelatif D, Fazel Y, Henry L, Wymer M. Global epidemiology of nonalcoholic fatty liver disease-meta-analytic assessment of revalence, incidence, and outcomes. Hepatology 2016;64:73–84. [DOI] [PubMed] [Google Scholar]
  • 2.Ray K The next global epidemic. Nat Rev Gastroenterol Hepatol 2013;10(11):621. [DOI] [PubMed] [Google Scholar]
  • 3.Ekstedt M, Nasr P, S K. Natural History of NAFLD/NASH. Curr Hepatol Rep 2017;16(4):391–397s. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Bertot L, Adams L. The Natural Course of Non-Alcoholic Fatty Liver Disease. Int J Mol Sci 2016;17(5):774. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Bataller R, Brenner DA. Liver fibrosis. J Clin Invest 2005;115(2):209–218. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Tsukada S, Parsons CJ, Rippe RA. Mechanisms of liver fibrosis. Clin Chim Acta 2006;364(1–2):33–60. [DOI] [PubMed] [Google Scholar]
  • 7.Friedman SL. Liver fibrosis -- from bench to bedside. J Hepatol 2003;38 Suppl 1:S38–53. [DOI] [PubMed] [Google Scholar]
  • 8.Poynard T, McHutchison J, Manns M, et al. Impact of pegylated interferon alfa-2b and ribavirin on liver fibrosis in patients with chronic hepatitis C. Gastroenterology 2002;122(5):1303–1313. [DOI] [PubMed] [Google Scholar]
  • 9.Dixon JB, Bhathal PS, Hughes NR, O’Brien PE. Nonalcoholic fatty liver disease: improvement in liver histological analysis with weight loss. Hepatology 2004;39(6):1647–1654. [DOI] [PubMed] [Google Scholar]
  • 10.Shi Y, Guo Q, Xia F, et al. MR elastography for the assessment of hepatic fibrosis in patients with chronic Hepatitis B infection: does histologic necroinflammation influence the measurement of hepatic stiffness? Radiology 2014;273(1):88–98. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Reeder S, Pineda A, Wen Z, et al. Iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL): application with fast spin-echo imaging. Magn Reson Med 2005;54:636–644. [DOI] [PubMed] [Google Scholar]
  • 12.Singh S, Venkatesh S, Loomba R, et al. Magnetic resonance elastography for staging liver fibrosis in non-alcoholic fatty liver disease: a diagnostic accuracy systematic review and individual participant data pooled analysis.. Eur Radiol 2016;26(5):31–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Dulai P, Sirlin C, Loomba R. MRI and MRE for non-invasive quantitative assessment of hepatic steatosis and fibrosis in NAFLD and NASH:clinical trials to clinical practice.. J Hepatol 2016;65:(1006–16). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Liang Y, Li D. Magnetic resonance elastography in staging liver fibrosis in non-alcoholic fatty liver disease: a pooled analysis of the diagnostic accuracy. BMC Gastroenterology 2020;20(89). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Loomba R, Wolfson T, Ang B, et al. Magnetic resonance elastography predicts advanced fibrosis in patients with nonalcoholic fatty liver disease: a prospective disease. Hepatology 2014;Epub ahead of print. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Singh S, Venkatesh S, Wang Z, et al. Diagnostic performance of magnetic resonance elastography in staging liver fibrosis: a systematic review and meta-analysis of individual participant data. Clin Gastroenterol Hepatol 2015;13(3):440–451. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Yin M, Glaser K, Talwalkar J, Chen J, Manduca A, Ehman R. Hepatic MR Elastography: Clinical Performance in a Series of 1377 Consecutive Examinations. Radiology 2016;278(1):114–124. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Shire NJ, Yin M, Chen J, et al. Test-retest repeatability of MR elastography for noninvasive liver fibrosis assessment in Hepatitis C. Journal of Magnetic Resonance Imaging 2011;34:947–955. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Imajo K, Kessoku T, Honda Y, et al. Magnetic resonance imaging more accurately classifies steatosis and fibrosis in patients with nonalcoholic fatty liver disease than transient elastography.. Gastroenterology 2016;50:626–637. [DOI] [PubMed] [Google Scholar]
  • 20.Park C, Nguyen P, Hernandez C, et al. Magnetic resonance elastography vs transient elastography in detection of fibrosis and noninvasive measurement of steatosis in patients with biopsy-proven nonalcoholic fatty liver disease.. Gastroenterology 2017;152(3):598–607. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Cui J, Heba E, Hernandez C, et al. Magnetic resonance elastography is superior to acoustic radiation force impulse for the diagnosis of fibrosis in patients with biopsy-proven nonalcoholic fatty liver disease: a prospective study. Hepatology 2016;63:453–461. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Yoon JH, Lee JM, Woo HS, et al. Staging of hepatic fibrosis: comparison of magnetic resonance elastography and shear wave elastography in the same individuals. Korean Journal of Radiology 2013;14(2):202–212. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Wagner M, Corcuera-Solano I, Lo G, et al. Technical failure of MR elastography examinations of the liver: experience from a large single-center study Radiology 2017;284(2):401–412. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Yin M, Talwalkar JA, Glaser KJ, et al. Assessment of hepatic fibrosis with magnetic resonance elastography. Clinical Gastroenterology and Hepatology 2007;5:1207–1213. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Kim D, Kim WR, Talwalkar JA, Kim HJ, Ehman RL. Advanced fibrosis in nonalcoholic fatty liver disease: noninvasive assessment with MR elastography. Radiology 2013;268(2):411–419. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Talwalkar JA, Yin M, Fidler JL, Sanderson SO, Kamath PS, Ehman RL. Magnetic resonance imaging of hepatic fibrosis: emerging clinical applications. Hepatology 2008;47(1):332–342. [DOI] [PubMed] [Google Scholar]
  • 27.Sigrist R, Liau J, Kaffas A, Chammas M, Willmann J. Ultrasound Elastography: Review of Techniques and Clinical Applications. Theranostics 2017;7(5):1303–1329. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Guglielmo F, Venkatesh S, Mitchell D. Liver MR Elastography Technique and Image Interpretation: Pearls and Pitfalls. Radiographics 2019;39(7):1983–2002. [DOI] [PubMed] [Google Scholar]
  • 29.Hines CDG, Bley TA, Lindstrom MJ, Reeder SB. Repeatability of magnetic resonance elastography for quantification of hepatic stiffness. Journal of Magnetic Resonance Imaging 2010;31:725–731. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Noureddin M, Lam J, Peterson M, et al. Utility of magnetic resonance imaging versus histology for quantifying changes in liver fat in nonalcoholic fatty liver disease trials.. Hepatology 2013;58(6):1930–1940. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Campo C, Hernando D, Schubert T, Bookwalter C, Van Pay A, Reeder S. Standarized Approach for ROI-Based Measurements of Proton Desnity Fat Fraction and R2* in the Liver. Am J Roentgenol 2017;209(3):592–603. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Kleiner D, Brunt E, Van Natta M, Behling C, Contos M, Cummings O. Design and validation of a histological scoring system for nonalcoholic fatty liver disease.. Hepatology 2005;41:1313–1321. [DOI] [PubMed] [Google Scholar]
  • 33.Committee QMEotLB. QIBA Profile: Magnetic Resonance Elastography of the Liver. QIBA 2018(Version profile stage: Consensus; ). [Google Scholar]
  • 34.Dzyubak B, Venkatesh S, Manduca A, Glaser J, Ehman R. Automated liver elasticity calculation for MR elastography. J Magn Reson Imag 2016;43(5):1055–1063. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Pluim J, Antoine Maintz J, Viergever M. Mutual information based registration of medical images: a survey. IEEE Trans Med Imag 2003;22(8):983–1004. [DOI] [PubMed] [Google Scholar]
  • 36.Castera L, Friedrich-Rust M, Loomba R. Noninvasive Assessment of Liver Disease in Patients with Nonalcoholic Fatty Liver Disease. Gastroenterology 2019;156:1264–1281. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Papagianni M, Sofogianni A, Tziomalos K. Non-invasive methods for the diagnosis of nonalcoholic fatty liver disease. World J Hepatol 2015;7(4):638–648. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Allen A, Shah V, Therneau T, et al. The Role of Three-Dimensional Magnetic Resonance Elastography in the Diagnosis of Nonalcoholic Steatohepatitis in Obese Patients Undergoing Bariatric Surgery. Hepatology 2019;0(0). [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES