- Email: [email protected]
Magnetic Resonance Elastography in Primary Sclerosing Cholangitis: Interobserver Agreement for Liver Stiffness Measurement with Manual and Automated Methods
ARTICLE IN PRESS
Original Investigation
Magnetic Resonance Elastography in Primary Sclerosing Cholangitis: Interobserver Agreement for Liver Stiffness Measurement with Manual and Automated Methods Safa Hoodeshenas, MD, Christopher L. Welle, MD, Patrick J. Navin, MB, BCh, BAO, Bogdan Dzyubak, PhD, John E. Eaton, MD, Richard L. Ehman, MD, Sudhakar K. Venkatesh, MD
Rationale and Objective: Primary sclerosing cholangitis, a chronic liver disease causes heterogeneous parenchymal changes and fibrosis. Liver stiffness measurement (LSM) with magnetic resonance Elastography (MRE) may be affected by this heterogeneous distribution. We evaluated interobserver agreement of LSM in primary sclerosing cholangitis (PSC) with manual and automated methods to study the influence of heterogeneous changes. Materials and Methods: A total of 79 consecutive patients with PSC who had a liver MRI and MRE formed the study group. Three readers with 1 3 years’ experience in MRE and a MRE expert (11 years’ experience) independently performed LSM. Each reader manually drew free hand (fROI) and average (aROI) on stiffness maps. Automatic liver elasticity calculation (ALEC) was used to generate automated LSM. The expert fROI was the reference standard. Correlation analysis and absolute intra-class correlation coefficient (ICC) analysis was performed. Results: LSM data of 79 livers and 315 sections were evaluated. There was excellent ICC between expert and reader fROIs (0.989, 95% confidence interval, and 0.985 0.993) and aROIs (0.971, 95% confidence interval, and 0.953 0.983) and ALEC (0.972, 0.957 0.982) with fROI performing better. The areas measured with fROIs and ALEC had moderate ICC with Expert fROI (0.64 and 0.56, respectively) whereas aROI area had a poor ICC of 0.12. Comparison of multiple methods showed significant differences in LSM between expert fROI and aROI of two readers and no significant differences for fROIs of all three readers. Conclusion: LSM with MRE in PSC patients shows excellent interobserver agreement with both fROI and aROI methods with better performance with fROI. fROI may therefore be preferred for LSM measurements in PSC. Key Words: Primary sclerosing cholangitis; Magnetic resonance elastography; Liver stiffness; Interobserver agreement; Reproducibility. © 2019 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.
INTRODUCTION
P
rimary Sclerosing Cholangitis (PSC) is a chronic cholestatic liver disease of unknown etiology and is characterized by multifocal and segmental biliary
Acad Radiol 2019; &:1 8 From the Department of Radiology, Mayo Clinic College of Medicine, Mayo Clinic, 200 First Street SW, Rochester, MN 55905 (S.H., C.L.W., P.J.N., B.D., R.L.E., S.K.V.); Division of Gastroenterology and Hepatology, Mayo Clinic College of Medicine, Rochester, Minnesota (J.E.E.). Received December 2, 2018; revised February 4, 2019; accepted February 5, 2019. Address correspondence to: S.K.V. e-mail: [email protected] © 2019 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.acra.2019.02.004
strictures. Chronic inflammation in PSC causes ductal inflammation and healing by liver fibrosis that ultimately progresses to cirrhosis and its associated complications. Moreover, patients with PSC have an increased risk of developing hepatobiliary and colorectal malignancies. Currently there is no effective treatment for PSC and many patients eventually need liver transplantation (1,2). PSC is diagnosed with demonstration of multisegmental biliary strictures on magnetic resonance cholangiography (MRC) or endoscopic retrograde cholangiography in patients with known inflammatory bowel disease and/ or absence of any other risk factor for chronic liver disease. PSC patients are monitored for progression of disease, which helps in making decisions on follow-up interval and timing of screening for complications of cirrhosis (3 6). Noninvasive 1
ARTICLE IN PRESS HOODESHENAS ET AL
methods, including ultrasound- and magnetic resonance-based elastography, are useful for detection and monitoring liver fibrosis and its progression (7 9). Magnetic resonance elastography (MRE) is useful in PSC as an additional sequence to the liver MRI and MRCs for detection and staging of liver fibrosis. MRE has proven to be an accurate, reproducible, and reliable technique for detection and staging of liver fibrosis in chronic liver diseases due to PSC as well as from other etiologies (3 6). Parenchymal involvement has an unusual heterogeneous distribution in PSC with peripheral, focal, and segmental regions of inflammation and fibrosis in early stages with relatively less affected central liver parenchyma. As the disease progresses, more fibrosis and atrophy occur in the periphery and, in response, large central regions of regenerative liver parenchyma develop. In the advanced stages, both peripheral and central regions including the regions of regeneration are affected and the entire liver becomes fibrotic or cirrhotic, similar to other causes of cirrhosis (10) (Fig 1). The early pathological changes in PSC may also involve only a few segments of the liver or a lobe with reactive hypertrophy of other segments and lobe. These changes are reflected on
Academic Radiology, Vol &, No &&, && 2019
MRE as peripheral and/or segmental regions of increased stiffness corresponding to inflammation and/or fibrosis; and central regions of normal or mildly increased liver stiffness that corresponds to regenerative parenchyma. This may add complexity to liver stiffness measurement (LSM), especially when only focal regions of increased or decreased stiffness are measured, leading to discrepancies during follow-up as different regions may be measured. Ultrasound-based elastography techniques typically measure small regions of interest (ROI) and therefore may result in poor reproducibility. LSM with MRE is performed by drawing one or more ROI in the typical four slices obtained. The mean stiffness from all the ROIs is taken as the representative LSM (6). The mean LSM gives an estimate of overall burden of the liver fibrosis and is the most common method for LSM with MRE. The most commonly used ROIs are either round or oval and one or more such ROI are placed in order to sample multiple regions of the liver. The other method is the freehand method in which an irregular or geographical ROI is drawn to include as large a part of liver as possible, resulting in better volume sampling. To remove or reduce measurement variability, it is recommended that either dedicated
Figure 1. MR Elastography stiffness maps in six different patients with PSC showing heterogeneous distribution of increased stiffness. The color scale (0 8) on top left corner of the image represents a stiffness scale of 0 8 kilopascals (kPa). The purple and blue colors represents the lower end of the scale and the orange-red colors represent the higher end of the scale. The distribution of increased liver stiffness is variable and tends to be peripheral in earlier stage of disease (A and B) with relatively normal or mildly increased stiffness in the central regions of the liver. This heterogeneous distribution of stiffness becomes more prominent or involves predominantly one segment or lobe more as disease progresses (D, E, and F). Note that in D, the posterior right lobe shows increased stiffness (arrow) compared to the anterior right lobe and left lobe. In E, there are focal regions of higher stiffness (arrowheads) within the peripheral rind of increased stiffness; and in F, there is more involvement of the posterior right lobe with focal higher stiffness in the periphery (arrow) and more involvement of the central liver in that lobe. In more advanced cases the central liver becomes involved C. (Color version of figure is available online.)
2
ARTICLE IN PRESS Academic Radiology, Vol &, No &&, && 2019
personnel, trained to draw contours under supervision of an experienced radiologist perform the measurements, or semiautomated or automated software be used if available (11 13). Given the heterogeneity of the stiffness in liver parenchyma in PSC, it is not known which ROI method has better reproducibility. Therefore, we evaluated reproducibility of LSM in PSC between three readers and an expert, with two types of ROI: freehand (fROI) and average (aROI). We also compared the readers’ LSM with automated liver elasticity calculation (ALEC) (14). MATERIALS AND METHODS This is an IRB-approved retrospective study and informed consent was waived for retrospective analysis. We searched our database for subjects with known PSC who also had a liver MRE between January 2016 and June 2017. The indications for liver MRI studies were either suspected PSC or follow-up evaluation of known PSC. All MRI and MRE studies were performed with standard clinical liver MRI and magnetic resonance cholangiopancreatography protocols, on either a 1.5T or 3T clinical MR scanner (Signa, GE Healthcare, Waukesha, Wisconsin). A passive driver (19 cm £ 1.5 cm) was placed against the anterior chest wall positioned over the right lobe of the liver, between the patient's chest/abdominal wall and the phased array coil. The active driver produced continuous acoustic vibrations at 60 Hz which were transmitted to the passive driver via a plastic tube to produce propagating shear waves in the liver. These waves were imaged with a modified phase contrast, gradient echo sequence. The sequence parameters were: repetition time/echo time = 100/25.6 ms; bandwidth = §31.25 kHz; flip angle = 30, field of view = 32 42 cm; matrix 256 £ 96; slice thickness 10 mm; gap 5 mm. For each patient, four MRE slices were obtained through the largest cross-section of the liver; a routine practice since the introduction of MRE for clinical use at our institution. These slices are typically placed in the largest cross-section of the liver, including the right and left lobe and in the upper third to half of the liver, avoiding the dome and the inferior part of the liver. Four slices through the largest cross-section of the liver provide a sufficiently large cross-section of liver parenchyma for sampling. MRE measures the mechanical properties of the liver and it is not dependent on scanner strength, provided that all other acquisition parameters including shear wave frequency, breath hold, and subject fasting status are similar. There is excellent test-retest repeatability of LSM across multiple manufacturer field strengths (1.5T vs. 3T), and pulse sequence combinations (15). An inversion algorithm installed in the scanner automatically generated wave images and stiffness maps in gray scale, color scale, and a color scale map with an overlaid confidence map. Three radiologists with 1 3 years’ experience in liver MRE independently performed LSM on an Advantage
MRE IN PRIMARY SCLEROSING CHOLANGITIS
Workstation (Advantage Windows, 2.0, GE). The readers drew fROI, avoiding liver edges, fissures, large vessels, wave interference and artifacts, using the magnitude image as reference, and adjusting the ROIs to fit within the valid regions of the confidence map. ROIs were drawn as large as possible and included regions of increased stiffness. Each individual reader also placed round or oval ROIs (aROI) within the confidence maps for each slice. The size and orientation of these aROIs were changed as required to fit within the confidence map and at the reader's discretion. The mean LSM and area of the ROI for each slice were recorded. The LSM for whole liver (average of all four slices) and for each ROI method (fROI and aROI) and the total area of liver interrogated (sum of ROIs) was also recorded. The same MRE data sets were processed using ALEC software (14) for automated LSMs (Fig 2). ALEC is an automated algorithm that can perform LSM in clinical liver MRE exams and was developed to address LSM variability among readers. ALEC performs the task in three stages initial tissue estimation, segmentation, and elasticity artifact removal (14). The initial tissue estimation step involves identifying liver tissue and separating it from surrounding tissues, particularly fat. The second step involves segmentation of the liver and the final step removes the artifacts from wave interference, partial volume, and low signal to noise ratio regions as those can significantly bias the calculation of mean liver stiffness. All these steps are performed on an offline independent workstation. An expert in MRE with more than 11 years’ experience independently drew freehand ROI (Expert fROI) and the LSM measurement was used as a reference standard. We used Expert fROI LSM as the reference standard because the expert had extensive experience with liver MRE (>5000 MREs) and understands the artifacts and wave interference that occurs in shear wave propagation. There is precedent for the use of an expert LSM as a reference standard in earlier validation studies (14,16,17). We were not able to use liver biopsy as it had not been performed or required for clinical management of the PSC. Statistical data was expressed as mean and standard deviation. Correlation between LSM by readers and ALEC was assessed with Pearson correlation analysis. Absolute intraclass correlation coefficient (ICC) analysis was performed for interobserver agreement, considering ICC values lower than 0.5 indicative of poor reliability, values between 0.5 and 0.75 indicative of moderate reliability, values between 0.75 and 0.9 indicative of good reliability, and values greater than 0.90 indicative of excellent reliability (18). We used a Bland-Altman plot with multiple measurements per subject-method to compare methods. The Bland-Altman plot is a graphical tool to compare two measurement techniques where the differences between the two methods are plotted against the averages of the two methods (19). This plot is useful to reveal relationships between the differences and averages in order to observe any systematic biases and to identify outliers in the study. Comparison of multiple 3
ARTICLE IN PRESS HOODESHENAS ET AL
Academic Radiology, Vol &, No &&, && 2019
Figure 2. Magnitude image (A) and MRE color stiffness maps with confidence overlay (B I) of a single slice through the liver in a patient with PSC. Freehand regions of interest drawn by an expert (B), automated liver elasticity calculation (C) and three independent readers (D F) are shown. Oval or rounded average ROIs drawn by three independent readers (G I). Note the differences in ROIs, both in shape and regions included. Magnitude image for reference.
methods is an extension of the Bland-Altman plot when there are more than two methods. For each method, the differences with a reference method are plotted against the values of the reference method (19). The procedure generates multiple bias plots in one single display to facilitate comparison of the different methods. Absolute percentage errors (APE) were evaluated and graphs were generated. The APE is calculated as 100 * [(y¡ref)/ref] where y is the observation and ref is the reference value. The median APE and 95% confidence interval (CI) are reported. The 95th percentile APE is interpreted as follows: the percentage difference between a measurement and the reference value is not expected to exceed this value with 95% certainty. As the study is on interobserver agreement, we used p < 0.01 as statistically significant. A higher p value could not be used as the study population was only 80 subjects. All 4
statistical analyses were performed with MedCalc Statistical Software version 12.7.7 (MedCalc Software bvba, Ostend, Belgium). RESULTS Eighty patients were eligible for the study during the recruitment period. One patient was excluded due to high iron content in the liver, which resulted in a failed MRE. The final study group was comprised of 79 patients (48 men and 31 women, mean age § SD, 44.24 years § 14.42, range 15 75 years). LSM data of 79 livers and 315 liver slices were available from both readers and ALEC. One slice was excluded due to obliquity of waves through the top section of the liver near the diaphragm. Mean LSMs ranged from 1.8 to 11.1 kPa. The total area of ROIs ranged from 75.2 to 541 cm2. Pearson
ARTICLE IN PRESS Academic Radiology, Vol &, No &&, && 2019
MRE IN PRIMARY SCLEROSING CHOLANGITIS
TABLE 1. Pearson Correlation Coefficient (r) Values Between LSM with Different ROI Methods* E-fROI E-fROI R1-fROI R2-fROI R3-fROI R1-aROI R2-aROI R3-aROI ALEC
0.989 0.993 0.992 0.972 0.980 0.989 0.983
R1-fROI
R2-fROI
ALEC
R1-aROI
R2-aROI
R3-aROI
R3-fROI
0.989
0.993 0.989
0.992 0.988 0.997
0.972 0.974 0.982 0.979
0.980 0.980 0.986 0.980 0.973
0.989 0.982 0.993 0.990 0.971 0.978
0.983 0.983 0.987 0.992 0.963 0.970 0.987
0.989 0.988 0.974 0.989 0.982 0.983
0.997 0.982 0.986 0.993 0.987
0.979 0.980 0.990 0.992
0.973 0.971 0.963
0.978 0.970
0.987
* All r values are statistically significant with p < 0.001.
correlation analysis showed excellent LSM correlation between the three readers, ALEC, and the expert (Table 1), whereas the correlation between areas measured was poor to good (Table 2). The intraclass correlation for both single measures (ICC = 0.98, 95% CI 0.97 0.985) and average measures (ICC = 0.99, 95% CI, 0.996 0.998) of LSM was excellent (Fig 2). With Expert fROI as reference, the APE of LSM ranged between 3.41% and 4.34% for fROI by readers, 4.9% for ALEC, and 3.72% and 5.45% for aROI by readers (Table 3). The Bland-Altman plots show more variation with aROIs in comparison to fROIs (Fig 3). The total area of ROIs evaluated by readers, ALEC, and the expert showed poor correlation for single measures (ICC = 0.24, 95% CI, 0.09 0.40) and moderate to good correlation for average measures (ICC = 0.71, 95% CI, 0.46 0.84). With Expert fROI as reference, the APE of the total area measured ranged between 10.7% and 26.4% for readers’ fROIs, 19.8% for ALEC, and 32.7% and 64.8% for readers’ aROIs (Table 3). The Bland-Altman plots show a high degree of variability for all the methods with Expert fROI as reference standard (Fig 4). DISCUSSION In this study we performed analysis of reproducibility of LSM with MRE in PSC patients among three independent readers drawing two different ROIs and automated software using LSM from an expert as a reference standard. The Bland-Altman Plot analysis showed that fROIs had better overall results compared to aROI drawn by independent readers. Our study
demonstrates excellent interobserver agreement between LSM using fROIs by readers and ALEC in PSC patients. The ICC was excellent even with modest correlation between areas of ROIs drawn by each reader, the expert, and ALEC. The larger ROIs performed better as these reduced sampling errors and therefore produced excellent correlation with LSM. We compared the ICC with the automated algorithm, which also selects a large ROI with artifact-free liver tissue, and showed excellent correlation. This shows robustness of measuring LSM with larger ROI. The ROIs drawn by independent readers and the ALEC methods included focal/segmental/lobar areas of increased stiffness. Therefore, drawing larger fROIs on MRE stiffness maps that include both large regions of increased stiffness and regions of lower stiffness ensure good reproducibility of LSM in patients with PSC. In our experience, mean LSM provides the most meaningful measurements for clinical decision-making. In PSC patients, the liver function remains maintained till the development of cirrhosis and portal hypertension or they develop a cholangiocarcinoma or other complications. The mean LSM that includes as large liver parenchyma as possible is more reflective of overall fibrosis burden and indirectly the residual functioning parenchyma and liver function as compared to measuring only focal regions of increased stiffness. Measuring the region of highest stiffness can also be misleading as it will invariably increase the stage of fibrosis when there may be large regions of regenerative parenchyma that maintains overall liver function. This can lead to discordant clinical and MRE findings. Therefore large ROIs that includes as much of liver parenchyma are preferred.
TABLE 2. Pearson Correlation Coefficient (r) Values Between Areas Measured with Different ROI Methods* E-fROI E-fROI R1-fROI R2-fROI R3-fROI ALEC R1-aROI R2-aROI R3-aROI
0.774 0.853 0.813 0.570 0.529 0.711 0.701
R1-fROI
R2-fROI
R3-fROI
ALEC
R1-aROI
R2-aROI
R3-aROI
0.774
0.853 0.730
0.813 0.696 0.856
0.570 0.582 0.646 0.526
0.529 0.541 0.614 0.509 0.434
0.711 0.622 0.815 0.700 0.611 0.506
0.701 0.576 0.735 0.836 0.438 0.431 0.601
0.730 0.696 0.582 0.541 0.622 0.576
0.856 0.646 0.614 0.815 0.735
0.526 0.509 0.700 0.836
0.434 0.611 0.438
0.506 0.431
0.601
* All r values are statistically significant with p < 0.001.
5
ARTICLE IN PRESS Academic Radiology, Vol &, No &&, && 2019
HOODESHENAS ET AL
TABLE 3. Median Absolute Percentage Errors (APE) with Bland-Altman Plots with Expert fROI as Reference Standard Method
Median APE for LSM (%)
95% CI
Median APE for Area (%)
95% CI
R1-fROI R2-fROI R3-fROI R1-aROI R2-aROI R3-aROI ALEC
4.09 3.41 4.34 3.72 5.11 5.45 4.89
2.85 2.58 3.03 2.88 4.06 4.37 3.78
26.41 10.13 10.71 32.65 62.17 64.78 19.83
22.25 7.86 9.16 28.19 59.88 63.57 13.66
Our study showed very high ICC values similar to the reported substantial to almost perfect ICCs ranging from 0.74 to 0.99 in prior studies (20 22).
4.74 4.18 5.36 4.88 6.51 6.61 6.53
30.77 12.75 13.96 37.39 63.52 67.54 25.45
The size of the ROIs did not influence the reproducibility and our results are similar to Runge et al. (3), who showed that the ICC between two readers was almost perfect
Figure 3. Bland-Altman plots showing comparison of LSM measurements of readers’ fROI and aROI methods and ALEC with reference standard of Expert fROI LSM. The Bland-Altman plot is a graphical tool to compare two measurement techniques where the differences between the two methods are plotted against the averages of the two techniques. This plot is useful to reveal the relationship between the differences and averages to observe any systematic biases and to identify outliers in the study. The x-axis represents the difference in LSM from the reference and the y-axis represents the mean of two methods. The horizontal black line represents the mean of differences, the hashed line represents the regression of differences, and grey dotted lines represent limits of agreement (§1.96 SD of the mean). Note the differences are very small for fROI and ALEC and the regression line is very close to the mean line. The regression line deviates more from the mean line for aROI particularly for R2 and R3.
6
ARTICLE IN PRESS Academic Radiology, Vol &, No &&, && 2019
MRE IN PRIMARY SCLEROSING CHOLANGITIS
Figure 4. Bland-Altman plots showing comparison of areas of fROI and aROI drawn by readers and ALEC with reference standard of Expert fROI. The Bland-Altman plot is a graphical tool to compare two measurements techniques where the differences between the two methods are plotted against the averages of the two techniques. This plot is useful to reveal relationship between the differences and averages to observe any systematic biases and to identify outliers in the study. The x-axis represents the differences in area from the reference and the y-axis represents the mean of two methods. The horizontal black line represents the mean of differences, the hashed line represents the regression of differences, and the grey dotted lines represent limits of agreement (§1.96 SD of the mean). Note large differences in the area covered by average ROI, particularly readers R2 and R3.
(rs = 0.97; ICC = 0.99) in 85 patients with chronic hepatitis B or C. It is expected that fibrosis distribution in chronic hepatitis B or C is likely to be more homogeneous as compared to PSC. However, with large ROIs the effect of heterogeneity is minimized as demonstrated in our study. Our study has some limitations. We did not have histological confirmation of fibrosis in all these patients; however, this was unavoidable as liver biopsies are not routinely performed in PSC patients and the diagnosis is established on clinical presentation, laboratory tests, and typical changes in the biliary tree demonstrated on MRCs or endoscopic retrograde cholangiopancreatography. However, this study was performed with a larger cohort in this uncommon disease and is probably one of the largest of PSC patients to be studied with MRE. Also, the comparison with ALEC software is unique to our study. ALEC is a useful method for clinical use as it
will eliminate reader bias. However, this automated method is not widely available and needs a separate independent workstation. In clinical practices where an expert reader is not available, one may consider the use of ALEC to avoid reader bias. Currently, efforts are underway to make this software widely available. We did not calculate the time required to draw the ROIs by the independent readers for comparison. Intuitively, however, fROIs take more time to draw as it takes practice to draw manually. In our clinical practice, ALEC takes about 5 6 minutes to perform LSM in one patient. In conclusion, LSM with MRE in PSC patients shows excellent interobserver agreement when large freehand ROIs are drawn to include regions of both increased and lower stiffness caused by the heterogeneous distribution of disease process. 7
ARTICLE IN PRESS HOODESHENAS ET AL
REFERENCES 1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
8
Eaton JE, Talwalkar JA, Lazaridis KN, et al. Pathogenesis of primary sclerosing cholangitis and advances in diagnosis and management. Gastroenterology 2013; 145:521–536 https://doi.org/10.1053/j.gastro.2013.06.052. Eaton JE, Dzyubak B, Venkatesh SK, et al. Performance of magnetic resonance elastography in primary sclerosing cholangitis. J Gastroenterol Hepatol 2016; 31:1184–1190 https://doi.org/10.1111/jgh.13263. Runge JH, Bohte AE, Verheij J, et al. Comparison of interobserver agreement of magnetic resonance elastography with histopathological staging of liver fibrosis. Abdom Imaging 2014; 39:283–290 https://doi.org/ 10.1007/s00261-013-0063-z. Singh S, Venkatesh SK, 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; 13:440–451. e6 https://doi.org/10.1016/j.cgh.2014.09.046. Venkatesh SK, Yin M, Ehman RL. Magnetic resonance elastography of liver: technique, analysis, and clinical applications. J Magn Reson Imaging 2013; 37:544–555 https://doi.org/10.1002/jmri.23731. Venkatesh SK, Yin M, Ehman RL. Magnetic resonance elastography of liver: clinical applications. J Comput Assist Tomogr 2013; 37:887–896 https://doi.org/10.1097/RCT.0000000000000032. Krawczyk M, Ligocka J, Ligocki M, et al. Does transient elastography correlate with liver fibrosis in patients with PSC? Laennec score-based analysis of explanted livers. Scand J Gastroenterol 2017; 52:1407–1412 https://doi.org/10.1080/00365521.2017.1370009. Mjelle AB, Mulabecirovic A, Hausken T, et al. Ultrasound and point shear wave elastography in livers of patients with primary sclerosing cholangitis. Ultrasound Med Biol 2016; 42:2146–2155 https://doi.org/10.1016/j. ultrasmedbio.2016.04.016. Righi S, Fiorini E, De Molo C, et al. ARFI elastography in patients with chronic autoimmune liver diseases: a preliminary study. J Ultrasound 2012; 15:226–231 https://doi.org/10.1016/j.jus.2012.10.002. Venkatesh SK, Wells ML, Miller FH, et al. Magnetic resonance elastography: beyond liver fibrosis-a case-based pictorial review. Abdom Radiol 2018; 43:1590–1611 https://doi.org/10.1007/s00261-017-1383-1.
Academic Radiology, Vol &, No &&, && 2019
11. Serai SD, Obuchowski NA, Venkatesh SK, et al. Repeatability of MR elastography of liver: a meta-analysis. Radiology 2017; 285:92–100 https://doi.org/10.1148/radiol.2017161398. 12. Shire NJ, Yin M, Chen J, et al. Test-retest repeatability of MR elastography for noninvasive liver fibrosis assessment in hepatitis C. J Magn Reson Imaging 2011; 34:947–955 https://doi.org/10.1002/jmri.22716. 13. Dzyubak B, Glaser K, Yin M, et al. Automated liver stiffness measurements with magnetic resonance elastography. J Magn Reson Imaging 2013; 38:371–379 https://doi.org/10.1002/jmri.23980. 14. Dzyubak B, Venkatesh SK, Manduca A, et al. Automated liver elasticity calculation for MR elastography. J Magn Reson Imaging 2016; 43:1055– 1063 https://doi.org/10.1002/jmri.25072. 15. Trout AT, Serai S, Mahley AD, et al. Liver stiffness measurements with MR elastography: agreement and repeatability across imaging systems, field strengths, and pulse sequences. Radiology 2016; 281:793–804 https://doi.org/10.1148/radiol.2016160209. 16. Dzyubak B, Venkatesh SK, Glaser K, et al. Stable automated segmentation of liver MR elastography images for clinical stiffness measurement. Proc SPIE Int Soc Opt Eng 2013: 8672 https://doi.org/10.1117/12.2006943. 17. Dzyubak B, Glaser KJ, Manduca A, et al. Automated liver elasticity calculation for 3D MRE. Proc SPIE Int Soc Opt Eng 2017: 10134 https://doi. org/10.1117/12.2254476. 18. Portney LG, Watkins MP. Foundations of Clinical Research: Applications to Practice: F.A. Davis Company/Publishers; 2015. 19. Krouwer JS. Why Bland-Altman plots should use X, not (Y+X)/2 when X is a reference method. Stat Med 2008; 27:778–780 https://doi.org/ 10.1002/sim.3086. 20. Lee DH, Lee JM, Han JK, et al. MR elastography of healthy liver parenchyma: Normal value and reliability of the liver stiffness value measurement. J Magn Reson Imaging 2013; 38:1215–1223 https://doi.org/10.1002/jmri.23958. 21. Rustogi R, Horowitz J, Harmath C, et al. Accuracy of MR elastography and anatomic MR imaging features in the diagnosis of severe hepatic fibrosis and cirrhosis. J Magn Reson Imaging 2012; 35:1356–1364 https://doi.org/10.1002/jmri.23585. 22. Venkatesh SK, Wang G, Lim SG, et al. Magnetic resonance elastography for the detection and staging of liver fibrosis in chronic hepatitis B. Eur Radiol 2014; 24:70–78 https://doi.org/10.1007/s00330-013-2978-8.
Original Investigation
Magnetic Resonance Elastography in Primary Sclerosing Cholangitis: Interobserver Agreement for Liver Stiffness Measurement with Manual and Automated Methods Safa Hoodeshenas, MD, Christopher L. Welle, MD, Patrick J. Navin, MB, BCh, BAO, Bogdan Dzyubak, PhD, John E. Eaton, MD, Richard L. Ehman, MD, Sudhakar K. Venkatesh, MD
Rationale and Objective: Primary sclerosing cholangitis, a chronic liver disease causes heterogeneous parenchymal changes and fibrosis. Liver stiffness measurement (LSM) with magnetic resonance Elastography (MRE) may be affected by this heterogeneous distribution. We evaluated interobserver agreement of LSM in primary sclerosing cholangitis (PSC) with manual and automated methods to study the influence of heterogeneous changes. Materials and Methods: A total of 79 consecutive patients with PSC who had a liver MRI and MRE formed the study group. Three readers with 1 3 years’ experience in MRE and a MRE expert (11 years’ experience) independently performed LSM. Each reader manually drew free hand (fROI) and average (aROI) on stiffness maps. Automatic liver elasticity calculation (ALEC) was used to generate automated LSM. The expert fROI was the reference standard. Correlation analysis and absolute intra-class correlation coefficient (ICC) analysis was performed. Results: LSM data of 79 livers and 315 sections were evaluated. There was excellent ICC between expert and reader fROIs (0.989, 95% confidence interval, and 0.985 0.993) and aROIs (0.971, 95% confidence interval, and 0.953 0.983) and ALEC (0.972, 0.957 0.982) with fROI performing better. The areas measured with fROIs and ALEC had moderate ICC with Expert fROI (0.64 and 0.56, respectively) whereas aROI area had a poor ICC of 0.12. Comparison of multiple methods showed significant differences in LSM between expert fROI and aROI of two readers and no significant differences for fROIs of all three readers. Conclusion: LSM with MRE in PSC patients shows excellent interobserver agreement with both fROI and aROI methods with better performance with fROI. fROI may therefore be preferred for LSM measurements in PSC. Key Words: Primary sclerosing cholangitis; Magnetic resonance elastography; Liver stiffness; Interobserver agreement; Reproducibility. © 2019 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.
INTRODUCTION
P
rimary Sclerosing Cholangitis (PSC) is a chronic cholestatic liver disease of unknown etiology and is characterized by multifocal and segmental biliary
Acad Radiol 2019; &:1 8 From the Department of Radiology, Mayo Clinic College of Medicine, Mayo Clinic, 200 First Street SW, Rochester, MN 55905 (S.H., C.L.W., P.J.N., B.D., R.L.E., S.K.V.); Division of Gastroenterology and Hepatology, Mayo Clinic College of Medicine, Rochester, Minnesota (J.E.E.). Received December 2, 2018; revised February 4, 2019; accepted February 5, 2019. Address correspondence to: S.K.V. e-mail: [email protected] © 2019 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.acra.2019.02.004
strictures. Chronic inflammation in PSC causes ductal inflammation and healing by liver fibrosis that ultimately progresses to cirrhosis and its associated complications. Moreover, patients with PSC have an increased risk of developing hepatobiliary and colorectal malignancies. Currently there is no effective treatment for PSC and many patients eventually need liver transplantation (1,2). PSC is diagnosed with demonstration of multisegmental biliary strictures on magnetic resonance cholangiography (MRC) or endoscopic retrograde cholangiography in patients with known inflammatory bowel disease and/ or absence of any other risk factor for chronic liver disease. PSC patients are monitored for progression of disease, which helps in making decisions on follow-up interval and timing of screening for complications of cirrhosis (3 6). Noninvasive 1
ARTICLE IN PRESS HOODESHENAS ET AL
methods, including ultrasound- and magnetic resonance-based elastography, are useful for detection and monitoring liver fibrosis and its progression (7 9). Magnetic resonance elastography (MRE) is useful in PSC as an additional sequence to the liver MRI and MRCs for detection and staging of liver fibrosis. MRE has proven to be an accurate, reproducible, and reliable technique for detection and staging of liver fibrosis in chronic liver diseases due to PSC as well as from other etiologies (3 6). Parenchymal involvement has an unusual heterogeneous distribution in PSC with peripheral, focal, and segmental regions of inflammation and fibrosis in early stages with relatively less affected central liver parenchyma. As the disease progresses, more fibrosis and atrophy occur in the periphery and, in response, large central regions of regenerative liver parenchyma develop. In the advanced stages, both peripheral and central regions including the regions of regeneration are affected and the entire liver becomes fibrotic or cirrhotic, similar to other causes of cirrhosis (10) (Fig 1). The early pathological changes in PSC may also involve only a few segments of the liver or a lobe with reactive hypertrophy of other segments and lobe. These changes are reflected on
Academic Radiology, Vol &, No &&, && 2019
MRE as peripheral and/or segmental regions of increased stiffness corresponding to inflammation and/or fibrosis; and central regions of normal or mildly increased liver stiffness that corresponds to regenerative parenchyma. This may add complexity to liver stiffness measurement (LSM), especially when only focal regions of increased or decreased stiffness are measured, leading to discrepancies during follow-up as different regions may be measured. Ultrasound-based elastography techniques typically measure small regions of interest (ROI) and therefore may result in poor reproducibility. LSM with MRE is performed by drawing one or more ROI in the typical four slices obtained. The mean stiffness from all the ROIs is taken as the representative LSM (6). The mean LSM gives an estimate of overall burden of the liver fibrosis and is the most common method for LSM with MRE. The most commonly used ROIs are either round or oval and one or more such ROI are placed in order to sample multiple regions of the liver. The other method is the freehand method in which an irregular or geographical ROI is drawn to include as large a part of liver as possible, resulting in better volume sampling. To remove or reduce measurement variability, it is recommended that either dedicated
Figure 1. MR Elastography stiffness maps in six different patients with PSC showing heterogeneous distribution of increased stiffness. The color scale (0 8) on top left corner of the image represents a stiffness scale of 0 8 kilopascals (kPa). The purple and blue colors represents the lower end of the scale and the orange-red colors represent the higher end of the scale. The distribution of increased liver stiffness is variable and tends to be peripheral in earlier stage of disease (A and B) with relatively normal or mildly increased stiffness in the central regions of the liver. This heterogeneous distribution of stiffness becomes more prominent or involves predominantly one segment or lobe more as disease progresses (D, E, and F). Note that in D, the posterior right lobe shows increased stiffness (arrow) compared to the anterior right lobe and left lobe. In E, there are focal regions of higher stiffness (arrowheads) within the peripheral rind of increased stiffness; and in F, there is more involvement of the posterior right lobe with focal higher stiffness in the periphery (arrow) and more involvement of the central liver in that lobe. In more advanced cases the central liver becomes involved C. (Color version of figure is available online.)
2
ARTICLE IN PRESS Academic Radiology, Vol &, No &&, && 2019
personnel, trained to draw contours under supervision of an experienced radiologist perform the measurements, or semiautomated or automated software be used if available (11 13). Given the heterogeneity of the stiffness in liver parenchyma in PSC, it is not known which ROI method has better reproducibility. Therefore, we evaluated reproducibility of LSM in PSC between three readers and an expert, with two types of ROI: freehand (fROI) and average (aROI). We also compared the readers’ LSM with automated liver elasticity calculation (ALEC) (14). MATERIALS AND METHODS This is an IRB-approved retrospective study and informed consent was waived for retrospective analysis. We searched our database for subjects with known PSC who also had a liver MRE between January 2016 and June 2017. The indications for liver MRI studies were either suspected PSC or follow-up evaluation of known PSC. All MRI and MRE studies were performed with standard clinical liver MRI and magnetic resonance cholangiopancreatography protocols, on either a 1.5T or 3T clinical MR scanner (Signa, GE Healthcare, Waukesha, Wisconsin). A passive driver (19 cm £ 1.5 cm) was placed against the anterior chest wall positioned over the right lobe of the liver, between the patient's chest/abdominal wall and the phased array coil. The active driver produced continuous acoustic vibrations at 60 Hz which were transmitted to the passive driver via a plastic tube to produce propagating shear waves in the liver. These waves were imaged with a modified phase contrast, gradient echo sequence. The sequence parameters were: repetition time/echo time = 100/25.6 ms; bandwidth = §31.25 kHz; flip angle = 30, field of view = 32 42 cm; matrix 256 £ 96; slice thickness 10 mm; gap 5 mm. For each patient, four MRE slices were obtained through the largest cross-section of the liver; a routine practice since the introduction of MRE for clinical use at our institution. These slices are typically placed in the largest cross-section of the liver, including the right and left lobe and in the upper third to half of the liver, avoiding the dome and the inferior part of the liver. Four slices through the largest cross-section of the liver provide a sufficiently large cross-section of liver parenchyma for sampling. MRE measures the mechanical properties of the liver and it is not dependent on scanner strength, provided that all other acquisition parameters including shear wave frequency, breath hold, and subject fasting status are similar. There is excellent test-retest repeatability of LSM across multiple manufacturer field strengths (1.5T vs. 3T), and pulse sequence combinations (15). An inversion algorithm installed in the scanner automatically generated wave images and stiffness maps in gray scale, color scale, and a color scale map with an overlaid confidence map. Three radiologists with 1 3 years’ experience in liver MRE independently performed LSM on an Advantage
MRE IN PRIMARY SCLEROSING CHOLANGITIS
Workstation (Advantage Windows, 2.0, GE). The readers drew fROI, avoiding liver edges, fissures, large vessels, wave interference and artifacts, using the magnitude image as reference, and adjusting the ROIs to fit within the valid regions of the confidence map. ROIs were drawn as large as possible and included regions of increased stiffness. Each individual reader also placed round or oval ROIs (aROI) within the confidence maps for each slice. The size and orientation of these aROIs were changed as required to fit within the confidence map and at the reader's discretion. The mean LSM and area of the ROI for each slice were recorded. The LSM for whole liver (average of all four slices) and for each ROI method (fROI and aROI) and the total area of liver interrogated (sum of ROIs) was also recorded. The same MRE data sets were processed using ALEC software (14) for automated LSMs (Fig 2). ALEC is an automated algorithm that can perform LSM in clinical liver MRE exams and was developed to address LSM variability among readers. ALEC performs the task in three stages initial tissue estimation, segmentation, and elasticity artifact removal (14). The initial tissue estimation step involves identifying liver tissue and separating it from surrounding tissues, particularly fat. The second step involves segmentation of the liver and the final step removes the artifacts from wave interference, partial volume, and low signal to noise ratio regions as those can significantly bias the calculation of mean liver stiffness. All these steps are performed on an offline independent workstation. An expert in MRE with more than 11 years’ experience independently drew freehand ROI (Expert fROI) and the LSM measurement was used as a reference standard. We used Expert fROI LSM as the reference standard because the expert had extensive experience with liver MRE (>5000 MREs) and understands the artifacts and wave interference that occurs in shear wave propagation. There is precedent for the use of an expert LSM as a reference standard in earlier validation studies (14,16,17). We were not able to use liver biopsy as it had not been performed or required for clinical management of the PSC. Statistical data was expressed as mean and standard deviation. Correlation between LSM by readers and ALEC was assessed with Pearson correlation analysis. Absolute intraclass correlation coefficient (ICC) analysis was performed for interobserver agreement, considering ICC values lower than 0.5 indicative of poor reliability, values between 0.5 and 0.75 indicative of moderate reliability, values between 0.75 and 0.9 indicative of good reliability, and values greater than 0.90 indicative of excellent reliability (18). We used a Bland-Altman plot with multiple measurements per subject-method to compare methods. The Bland-Altman plot is a graphical tool to compare two measurement techniques where the differences between the two methods are plotted against the averages of the two methods (19). This plot is useful to reveal relationships between the differences and averages in order to observe any systematic biases and to identify outliers in the study. Comparison of multiple 3
ARTICLE IN PRESS HOODESHENAS ET AL
Academic Radiology, Vol &, No &&, && 2019
Figure 2. Magnitude image (A) and MRE color stiffness maps with confidence overlay (B I) of a single slice through the liver in a patient with PSC. Freehand regions of interest drawn by an expert (B), automated liver elasticity calculation (C) and three independent readers (D F) are shown. Oval or rounded average ROIs drawn by three independent readers (G I). Note the differences in ROIs, both in shape and regions included. Magnitude image for reference.
methods is an extension of the Bland-Altman plot when there are more than two methods. For each method, the differences with a reference method are plotted against the values of the reference method (19). The procedure generates multiple bias plots in one single display to facilitate comparison of the different methods. Absolute percentage errors (APE) were evaluated and graphs were generated. The APE is calculated as 100 * [(y¡ref)/ref] where y is the observation and ref is the reference value. The median APE and 95% confidence interval (CI) are reported. The 95th percentile APE is interpreted as follows: the percentage difference between a measurement and the reference value is not expected to exceed this value with 95% certainty. As the study is on interobserver agreement, we used p < 0.01 as statistically significant. A higher p value could not be used as the study population was only 80 subjects. All 4
statistical analyses were performed with MedCalc Statistical Software version 12.7.7 (MedCalc Software bvba, Ostend, Belgium). RESULTS Eighty patients were eligible for the study during the recruitment period. One patient was excluded due to high iron content in the liver, which resulted in a failed MRE. The final study group was comprised of 79 patients (48 men and 31 women, mean age § SD, 44.24 years § 14.42, range 15 75 years). LSM data of 79 livers and 315 liver slices were available from both readers and ALEC. One slice was excluded due to obliquity of waves through the top section of the liver near the diaphragm. Mean LSMs ranged from 1.8 to 11.1 kPa. The total area of ROIs ranged from 75.2 to 541 cm2. Pearson
ARTICLE IN PRESS Academic Radiology, Vol &, No &&, && 2019
MRE IN PRIMARY SCLEROSING CHOLANGITIS
TABLE 1. Pearson Correlation Coefficient (r) Values Between LSM with Different ROI Methods* E-fROI E-fROI R1-fROI R2-fROI R3-fROI R1-aROI R2-aROI R3-aROI ALEC
0.989 0.993 0.992 0.972 0.980 0.989 0.983
R1-fROI
R2-fROI
ALEC
R1-aROI
R2-aROI
R3-aROI
R3-fROI
0.989
0.993 0.989
0.992 0.988 0.997
0.972 0.974 0.982 0.979
0.980 0.980 0.986 0.980 0.973
0.989 0.982 0.993 0.990 0.971 0.978
0.983 0.983 0.987 0.992 0.963 0.970 0.987
0.989 0.988 0.974 0.989 0.982 0.983
0.997 0.982 0.986 0.993 0.987
0.979 0.980 0.990 0.992
0.973 0.971 0.963
0.978 0.970
0.987
* All r values are statistically significant with p < 0.001.
correlation analysis showed excellent LSM correlation between the three readers, ALEC, and the expert (Table 1), whereas the correlation between areas measured was poor to good (Table 2). The intraclass correlation for both single measures (ICC = 0.98, 95% CI 0.97 0.985) and average measures (ICC = 0.99, 95% CI, 0.996 0.998) of LSM was excellent (Fig 2). With Expert fROI as reference, the APE of LSM ranged between 3.41% and 4.34% for fROI by readers, 4.9% for ALEC, and 3.72% and 5.45% for aROI by readers (Table 3). The Bland-Altman plots show more variation with aROIs in comparison to fROIs (Fig 3). The total area of ROIs evaluated by readers, ALEC, and the expert showed poor correlation for single measures (ICC = 0.24, 95% CI, 0.09 0.40) and moderate to good correlation for average measures (ICC = 0.71, 95% CI, 0.46 0.84). With Expert fROI as reference, the APE of the total area measured ranged between 10.7% and 26.4% for readers’ fROIs, 19.8% for ALEC, and 32.7% and 64.8% for readers’ aROIs (Table 3). The Bland-Altman plots show a high degree of variability for all the methods with Expert fROI as reference standard (Fig 4). DISCUSSION In this study we performed analysis of reproducibility of LSM with MRE in PSC patients among three independent readers drawing two different ROIs and automated software using LSM from an expert as a reference standard. The Bland-Altman Plot analysis showed that fROIs had better overall results compared to aROI drawn by independent readers. Our study
demonstrates excellent interobserver agreement between LSM using fROIs by readers and ALEC in PSC patients. The ICC was excellent even with modest correlation between areas of ROIs drawn by each reader, the expert, and ALEC. The larger ROIs performed better as these reduced sampling errors and therefore produced excellent correlation with LSM. We compared the ICC with the automated algorithm, which also selects a large ROI with artifact-free liver tissue, and showed excellent correlation. This shows robustness of measuring LSM with larger ROI. The ROIs drawn by independent readers and the ALEC methods included focal/segmental/lobar areas of increased stiffness. Therefore, drawing larger fROIs on MRE stiffness maps that include both large regions of increased stiffness and regions of lower stiffness ensure good reproducibility of LSM in patients with PSC. In our experience, mean LSM provides the most meaningful measurements for clinical decision-making. In PSC patients, the liver function remains maintained till the development of cirrhosis and portal hypertension or they develop a cholangiocarcinoma or other complications. The mean LSM that includes as large liver parenchyma as possible is more reflective of overall fibrosis burden and indirectly the residual functioning parenchyma and liver function as compared to measuring only focal regions of increased stiffness. Measuring the region of highest stiffness can also be misleading as it will invariably increase the stage of fibrosis when there may be large regions of regenerative parenchyma that maintains overall liver function. This can lead to discordant clinical and MRE findings. Therefore large ROIs that includes as much of liver parenchyma are preferred.
TABLE 2. Pearson Correlation Coefficient (r) Values Between Areas Measured with Different ROI Methods* E-fROI E-fROI R1-fROI R2-fROI R3-fROI ALEC R1-aROI R2-aROI R3-aROI
0.774 0.853 0.813 0.570 0.529 0.711 0.701
R1-fROI
R2-fROI
R3-fROI
ALEC
R1-aROI
R2-aROI
R3-aROI
0.774
0.853 0.730
0.813 0.696 0.856
0.570 0.582 0.646 0.526
0.529 0.541 0.614 0.509 0.434
0.711 0.622 0.815 0.700 0.611 0.506
0.701 0.576 0.735 0.836 0.438 0.431 0.601
0.730 0.696 0.582 0.541 0.622 0.576
0.856 0.646 0.614 0.815 0.735
0.526 0.509 0.700 0.836
0.434 0.611 0.438
0.506 0.431
0.601
* All r values are statistically significant with p < 0.001.
5
ARTICLE IN PRESS Academic Radiology, Vol &, No &&, && 2019
HOODESHENAS ET AL
TABLE 3. Median Absolute Percentage Errors (APE) with Bland-Altman Plots with Expert fROI as Reference Standard Method
Median APE for LSM (%)
95% CI
Median APE for Area (%)
95% CI
R1-fROI R2-fROI R3-fROI R1-aROI R2-aROI R3-aROI ALEC
4.09 3.41 4.34 3.72 5.11 5.45 4.89
2.85 2.58 3.03 2.88 4.06 4.37 3.78
26.41 10.13 10.71 32.65 62.17 64.78 19.83
22.25 7.86 9.16 28.19 59.88 63.57 13.66
Our study showed very high ICC values similar to the reported substantial to almost perfect ICCs ranging from 0.74 to 0.99 in prior studies (20 22).
4.74 4.18 5.36 4.88 6.51 6.61 6.53
30.77 12.75 13.96 37.39 63.52 67.54 25.45
The size of the ROIs did not influence the reproducibility and our results are similar to Runge et al. (3), who showed that the ICC between two readers was almost perfect
Figure 3. Bland-Altman plots showing comparison of LSM measurements of readers’ fROI and aROI methods and ALEC with reference standard of Expert fROI LSM. The Bland-Altman plot is a graphical tool to compare two measurement techniques where the differences between the two methods are plotted against the averages of the two techniques. This plot is useful to reveal the relationship between the differences and averages to observe any systematic biases and to identify outliers in the study. The x-axis represents the difference in LSM from the reference and the y-axis represents the mean of two methods. The horizontal black line represents the mean of differences, the hashed line represents the regression of differences, and grey dotted lines represent limits of agreement (§1.96 SD of the mean). Note the differences are very small for fROI and ALEC and the regression line is very close to the mean line. The regression line deviates more from the mean line for aROI particularly for R2 and R3.
6
ARTICLE IN PRESS Academic Radiology, Vol &, No &&, && 2019
MRE IN PRIMARY SCLEROSING CHOLANGITIS
Figure 4. Bland-Altman plots showing comparison of areas of fROI and aROI drawn by readers and ALEC with reference standard of Expert fROI. The Bland-Altman plot is a graphical tool to compare two measurements techniques where the differences between the two methods are plotted against the averages of the two techniques. This plot is useful to reveal relationship between the differences and averages to observe any systematic biases and to identify outliers in the study. The x-axis represents the differences in area from the reference and the y-axis represents the mean of two methods. The horizontal black line represents the mean of differences, the hashed line represents the regression of differences, and the grey dotted lines represent limits of agreement (§1.96 SD of the mean). Note large differences in the area covered by average ROI, particularly readers R2 and R3.
(rs = 0.97; ICC = 0.99) in 85 patients with chronic hepatitis B or C. It is expected that fibrosis distribution in chronic hepatitis B or C is likely to be more homogeneous as compared to PSC. However, with large ROIs the effect of heterogeneity is minimized as demonstrated in our study. Our study has some limitations. We did not have histological confirmation of fibrosis in all these patients; however, this was unavoidable as liver biopsies are not routinely performed in PSC patients and the diagnosis is established on clinical presentation, laboratory tests, and typical changes in the biliary tree demonstrated on MRCs or endoscopic retrograde cholangiopancreatography. However, this study was performed with a larger cohort in this uncommon disease and is probably one of the largest of PSC patients to be studied with MRE. Also, the comparison with ALEC software is unique to our study. ALEC is a useful method for clinical use as it
will eliminate reader bias. However, this automated method is not widely available and needs a separate independent workstation. In clinical practices where an expert reader is not available, one may consider the use of ALEC to avoid reader bias. Currently, efforts are underway to make this software widely available. We did not calculate the time required to draw the ROIs by the independent readers for comparison. Intuitively, however, fROIs take more time to draw as it takes practice to draw manually. In our clinical practice, ALEC takes about 5 6 minutes to perform LSM in one patient. In conclusion, LSM with MRE in PSC patients shows excellent interobserver agreement when large freehand ROIs are drawn to include regions of both increased and lower stiffness caused by the heterogeneous distribution of disease process. 7
ARTICLE IN PRESS HOODESHENAS ET AL
REFERENCES 1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
8
Eaton JE, Talwalkar JA, Lazaridis KN, et al. Pathogenesis of primary sclerosing cholangitis and advances in diagnosis and management. Gastroenterology 2013; 145:521–536 https://doi.org/10.1053/j.gastro.2013.06.052. Eaton JE, Dzyubak B, Venkatesh SK, et al. Performance of magnetic resonance elastography in primary sclerosing cholangitis. J Gastroenterol Hepatol 2016; 31:1184–1190 https://doi.org/10.1111/jgh.13263. Runge JH, Bohte AE, Verheij J, et al. Comparison of interobserver agreement of magnetic resonance elastography with histopathological staging of liver fibrosis. Abdom Imaging 2014; 39:283–290 https://doi.org/ 10.1007/s00261-013-0063-z. Singh S, Venkatesh SK, 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; 13:440–451. e6 https://doi.org/10.1016/j.cgh.2014.09.046. Venkatesh SK, Yin M, Ehman RL. Magnetic resonance elastography of liver: technique, analysis, and clinical applications. J Magn Reson Imaging 2013; 37:544–555 https://doi.org/10.1002/jmri.23731. Venkatesh SK, Yin M, Ehman RL. Magnetic resonance elastography of liver: clinical applications. J Comput Assist Tomogr 2013; 37:887–896 https://doi.org/10.1097/RCT.0000000000000032. Krawczyk M, Ligocka J, Ligocki M, et al. Does transient elastography correlate with liver fibrosis in patients with PSC? Laennec score-based analysis of explanted livers. Scand J Gastroenterol 2017; 52:1407–1412 https://doi.org/10.1080/00365521.2017.1370009. Mjelle AB, Mulabecirovic A, Hausken T, et al. Ultrasound and point shear wave elastography in livers of patients with primary sclerosing cholangitis. Ultrasound Med Biol 2016; 42:2146–2155 https://doi.org/10.1016/j. ultrasmedbio.2016.04.016. Righi S, Fiorini E, De Molo C, et al. ARFI elastography in patients with chronic autoimmune liver diseases: a preliminary study. J Ultrasound 2012; 15:226–231 https://doi.org/10.1016/j.jus.2012.10.002. Venkatesh SK, Wells ML, Miller FH, et al. Magnetic resonance elastography: beyond liver fibrosis-a case-based pictorial review. Abdom Radiol 2018; 43:1590–1611 https://doi.org/10.1007/s00261-017-1383-1.
Academic Radiology, Vol &, No &&, && 2019
11. Serai SD, Obuchowski NA, Venkatesh SK, et al. Repeatability of MR elastography of liver: a meta-analysis. Radiology 2017; 285:92–100 https://doi.org/10.1148/radiol.2017161398. 12. Shire NJ, Yin M, Chen J, et al. Test-retest repeatability of MR elastography for noninvasive liver fibrosis assessment in hepatitis C. J Magn Reson Imaging 2011; 34:947–955 https://doi.org/10.1002/jmri.22716. 13. Dzyubak B, Glaser K, Yin M, et al. Automated liver stiffness measurements with magnetic resonance elastography. J Magn Reson Imaging 2013; 38:371–379 https://doi.org/10.1002/jmri.23980. 14. Dzyubak B, Venkatesh SK, Manduca A, et al. Automated liver elasticity calculation for MR elastography. J Magn Reson Imaging 2016; 43:1055– 1063 https://doi.org/10.1002/jmri.25072. 15. Trout AT, Serai S, Mahley AD, et al. Liver stiffness measurements with MR elastography: agreement and repeatability across imaging systems, field strengths, and pulse sequences. Radiology 2016; 281:793–804 https://doi.org/10.1148/radiol.2016160209. 16. Dzyubak B, Venkatesh SK, Glaser K, et al. Stable automated segmentation of liver MR elastography images for clinical stiffness measurement. Proc SPIE Int Soc Opt Eng 2013: 8672 https://doi.org/10.1117/12.2006943. 17. Dzyubak B, Glaser KJ, Manduca A, et al. Automated liver elasticity calculation for 3D MRE. Proc SPIE Int Soc Opt Eng 2017: 10134 https://doi. org/10.1117/12.2254476. 18. Portney LG, Watkins MP. Foundations of Clinical Research: Applications to Practice: F.A. Davis Company/Publishers; 2015. 19. Krouwer JS. Why Bland-Altman plots should use X, not (Y+X)/2 when X is a reference method. Stat Med 2008; 27:778–780 https://doi.org/ 10.1002/sim.3086. 20. Lee DH, Lee JM, Han JK, et al. MR elastography of healthy liver parenchyma: Normal value and reliability of the liver stiffness value measurement. J Magn Reson Imaging 2013; 38:1215–1223 https://doi.org/10.1002/jmri.23958. 21. Rustogi R, Horowitz J, Harmath C, et al. Accuracy of MR elastography and anatomic MR imaging features in the diagnosis of severe hepatic fibrosis and cirrhosis. J Magn Reson Imaging 2012; 35:1356–1364 https://doi.org/10.1002/jmri.23585. 22. Venkatesh SK, Wang G, Lim SG, et al. Magnetic resonance elastography for the detection and staging of liver fibrosis in chronic hepatitis B. Eur Radiol 2014; 24:70–78 https://doi.org/10.1007/s00330-013-2978-8.