Assessment of Advanced Hepatic MR Elastography Methods for Susceptibility Artifact Suppression in Clinical Patients

Abstract

Purpose:

To assess the success rate, image quality, and the ability to stage liver fibrosis of a standard 2D gradient-recalled echo (GRE) and four different spin-echo (SE) magnetic resonance elastography (MRE) sequences in patients with different liver iron concentrations.

Materials and Methods:

A total of 332 patients who underwent 3T MRE examinations that included liver fat and iron quantification were enrolled, including 136 patients with all five MRE techniques. Thirty-four patients had biopsy results for fibrosis staging. The liver stiffness, region of interest area, image quality, and success rate of the five sequences were compared in 115/136 patients. The area under the receiver operating characteristic curves (AUCs) and the accuracies for diagnosing early-stage fibrosis and advanced fibrosis were compared. The effect of BMI (body mass index), the R2* relaxation time, and fat fraction on the image quality and liver stiffness measurements were analyzed.

Results:

The success rates were significantly higher in the four SE sequences (99.1–100%) compared with GRE MRE (85.3%) (all P < 0.001). There were significant differences of the mean ROI area between every pair of sequences (all P<0.0001). There were no significant differences in the AUC of the five MRE sequences for discriminating advanced fibrosis (10 P-values ranging from 0.2410–0.9171). R2* had a significant Effect on the success rate and image quality for the noniron 2D echo-planar imaging (EPI), 3D EPI and 2D GRE (all P < 0.001) sequences. BMI had a significant Effect on the iron 2D EPI (P = 0.0230) and iron 2D SE (P = 0.0040) sequences.

Conclusion:

All five techniques showed good diagnostic performance in staging liver fibrosis. The SE MRE sequences had higher success rates and better image quality than GRE MRE in 3T clinical hepatic imaging.

Magnetic resonance elastography (MRE) has the potenial to improve the characterization of diffuse and focal liver diseases by providing quantitative information about tissue viscoelasticity.^1–10 MRE has been shown to be a robust noninvasive technique for the detection and staging of hepatic fibrosis, particularly in early stages not well evaluated with other techniques, and has the potential to reduce or eliminate the need for invasive liver biopsies.¹¹ Currently, 2D gradient-recalled echo (GRE) MRE is the most commonly used technique for liver fibrosis assessment and it has been shown to have a high degree of accuracy and a high success rate.^8,10 Compared to ultrasound-based elastography, MRE images a larger volume of the liver, measures the stiff ness of deeper liver tissue, and is more successful when imaging obese patients and patients with ascites.^8,10,12

Despite these advantages, GRE MRE can fail in patients with moderate to severe liver iron overload due to reduced signal from the liver.^8,12 The echo time (TE) of MRE pulse sequences is prolonged due to the addition of motion-encoding gradients (MEGs), resulting in a low signal-to-noise ratio (SNR) in patients with high liver iron concentration (LIC) due to R2*-related signal decay. Yin et al⁸ reported that in routine clinical use, hepatic MRE had a 5.6% failure rate, with most failures being caused by low SNR likely due to liver iron overload.

One common technique to minimize the impact of the MEGs on the TE is to use MEGs with a duration shorter than the period of the mechanical motion.^13,1 However, since reducing the duration of the MEGs also reduces motion sensitivity, there are practical limits to how short the MEGs can be for a specific application. While decreasing the TE of GRE MRE can improve SNR,^13,15 for patients with significant hepatic iron overload this may not improve the SNR enough for a successful exam. However, if a spin-echo (SE) MRE pulse sequence with a short TE is used, the decay of the transverse magnetization becomes more dependent on T2 rather than T*₂ and it may be possible to maintain adequate image SNR even for patients with high LIC.^16–18 Excessive accumulation of iron and fat often occur in chronic liver disease and likely have synergistic liver injury mechanisms.¹⁹ MR quantification, using techniques such as the iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL-IQ), is now considered the method of choice for assessing liver iron and fat content because of its noninvasive nature and its wide availability for confirming a diagnosis, determining disease severity, and monitoring therapy with high sensitivity, specificity, and positive and negative predictive values.^20–22

The hypothesis of this study was that SE MRE sequences can improve the success rate and image quality of 3T liver MRE exams while preserving the accuracy of the stiffness measurements when compared with the standard 2D GRE MRE technique. The aim of this study was to retrospectively assess the quality and success rate of 2D GRE MRE and four different SE MRE sequences in patients with different LIC, using the R2* value derived from IDEAL-IQ as a surrogate for LIC.

MATERIALS AND METHODS

Subjects

With Institutional Review Board approval and written informed consent, 361 consecutive patients were enrolled in this study and underwent liver MRExaminations between October 2014 and April 2015 that included MRE and ID EAL-IQ. The clinical indications of chronic liver disease for these patients are listed in Table 1.

TABLE 1.

Clinical Indications for Liver MR Elastography Patients

Clinical indication	No. of cases	Percent (%)
HBV	208	62.7
Focal Lesion	56	16.9
Normal Liver in Final Diagnosis	11	3.3
HCV	7	2.1
Steatohepatitis	7	2.1
ALD	5	1.5
Biliary Obstruction	4	1.2
FHN	4	1.2
Hepatolithiasis	4	1.2
Cryptogenic	3	0.9
HBV+ALD	3	0.9
Congenital Biliary Dilatation	2	0.6
DIH	2	0.6
HBV+HCV	2	0.6
HBV+HEV	2	0.6
HBV + Steatohepatitis	2	0.6
Abnormal Liver Function Tests	1	0.3
AIH	1	0.3
BCS	1	0.3
Biliary Cirrhosis	1	0.3
Congestive Hepatopathy	1	0.3
HAV+HBV	1	0.3
HBV+DIH	1	0.3
HBV + HCV + G6PD	1	0.3
Parasitic Disease of the Liver	1	0.3
Portal Vein Thrombosis	1	0.3
Total	332	100

H {A,B,C,E}V: hepatitis {A,B,C,E} viral infection; ALD: alcoholic liver disease; FHN: focal nodular hyperplasia; DIH: drug-induced hepatotoxicity; BCS: Budd-Chiari syndrome; G6PD: Glucose-6-phosphate dehydrogenase deficiency.

MR Elastography

IMAGING.

All patients were imaged using a 3T clinical scanner (Discovery MR 750, GE Healthcare, Milwaukee, WI) with an 8-channel, phased-array, torso coil. The patients underwent the IDEAL-IQ scan for R2 * and fat-fraction measurements, and also underwent four or five different MRE acquisitions to measure hepatic stiffness. In the context of this study, we will refer to the five MRE techniques as: 1) “3D echo-planar imaging (EPI) MRE”: amoderate-TE SE-EPI protocol for 3D MRE (TE = 52 msec); 2) “noniron 2D EPIMRE”: a similar moderate-TE SE-EPI protocol for 2D MRE (TE = 52 msec)²³; 3) “iron 2D SE MRE”: a short-TE, SE, non-EPI protocol optimized for 2D MRE of patients with iron overload (TE = 10–12 msec); 4) “iron 2D EPIMRE”: a similar short-TE, SE-EPI protocol optimized for 2D MRE of patients with iron overload (TE = 10–12 msec)^16,17; and 5) “2D GRE MRE”: the standard GRE MRE protocol for 2D MRE (TE = 22 msec).² The five MRE sequences were not respiratory gated and were performed in one t ofour end-expiration breath-holds each, with scan times ranging from 16–136 seconds, depending on the patient size, weight, and ability to perform breath-holds. The 2D MRE scans for each subject were performed using the same four-slice slice locations (axial images through the widest part of the liver), while the 3D MRE scan was performed with 32, thinner, axial slices covering the center of the liver.

The primary MRE parameters are summarized in Table 2. The choice of parameters for this study was based on a number of criteria. The GRE MRE parameters were based on the standard implementation of four-slice, four-time-offset, clinical MRE at 1.5T which utilizes 1^st-gradient-moment-nulled (GMN) imaging gradients; a 16.7-msec, 1^st GMN, Z-direction MEG; and is typically performed in four, 14-second breath-holds (BH s). The non-iron 2D EPI MRE parameters were chosen to provide a single-BH, moderate-TE, SE alternative to GRE MRE with similar slice coverage and motion and flow sensitivity, and utilized a single-shot EPI acquisition; ^1st GMN imaging gradients; 6.45-msec, O^th-GMN, Z-direction MEG s on each side of the SE refocusing pulse; and was performed in one, 16-second BH. The 3D EPI MRE parameters were chosen to be similar to the noniron 2D EPI sequence while allowing for the measurement of a significant volume of the liver with thinner slices and all three motion-encoding directions in a time similar to the original GRE MRE sequence. This sequence acquires three time offsets, 32 3.6-mm slices, and all three motion-encoding directions in three, 21-second BHs (or six, 11-second BHs, if needed). The iron 2D SE and EPI parameters were designed with the primary requirement being a short TE of 10–12 msec to compensate for the short T2 of iron-overloaded liver. For the iron 2D EPI sequence, this was accomplished by only using O^th-GMN imaging gradients; a 2-msec, single-lobe, trapezoidal, Z-direction MEG on each side of the refocusing pulse; and eight EPI shots. To keep the BH time reasonable, the repetition time (TR) had to be reduced to 250–333 msec, and the scans were performed in either one, 24-second BH or two, 15-second BH s. Similar parameters were chosen for the iron 2D SE (non-EPI) sequence as well, but because it is a slower acquisition, it was typically performed in four BHs that were 17.7 −34.0 seconds long, depending on how the size of the patient affected the minimum TR due to SAR (RF specific absorption rate) limitations.

TABLE 2.

Parameters for the 5 MRE Sequences

Parameters	3D EPI MRE	Non-iron 2D EPI MRE	Iron 2D EPI MRE	Iron 2D SEMRE	2D GRE MRE
Acquisition matrix	80×80	80×80	80×80	128×64	256×64
TE (msec)	51.7–52.8	51.4–52.5	10.2–11.9	10.4–12.4	22
TR (msec)	1334	1000–1067	250–333	167–333	50 ms
EPI Shots	1	1	8	NA	NA
# of Breath Holds × Breath Hold Time (s)	3 × 21	1 × 16	(1 × 24) or (2 × 15)	4 × (17.7–34.0)	4 × (10.5–13.7)
FOV (cm)	38.7–44.8	38–44	38–44	38–44	38–44
# of Slices × Slice Thickness (mm)	32 × 3.6	4 × 10	4 × 10	4 × 10	4 × 10
Parallel Imaging	2	2	2	1	2
Acceleration Factor
Phase Offsets	3	4	4	4	4
Receiver Bandwidth	500 kHz	500 kHz	500 kHz	125 kHz	62.5 kHz
% FOV in	100	75–100	75–100	75–100	75–100
Phase-Encoding
Direction
Flow Compensation	Yes	Yes	No	No	Yes
Superior-Inferior Spatial	Yes	Yes	Yes	Yes	Yes
Presaturation Bands
Motion Sensitivity (μm/radian)	7.86	7.86	20.8–23.8	13.1	10.2
Motion-Encoding Directions	±X,Y,Z	±Z	±Z	±Z	±Z

TE: echo time; TR: repetition time; FOV: field of view; 2D: two-dimensional; 3D: three-dimensional; EPI: echo-planar imaging; GRE: gradient-recalled echo; MRE: magnetic resonance elastography; SE: spin echo.

To reduce potential physiological confounding factors, patients were instructed to fast for a minim um of 4 hours before the MRE exams. A pneumatic passive driver was placed over the lower chest and upper abdomen, over the right lobe of the liver, at the level of the xiphisternum. Continuous mechanical vibrations at 60 Hz were generated using an active acoustic driver located outside of the scan room and were transmitted through PVC tubing to the passive driver to produce shear waves in the liver.² The MRE magnitude and phase images were processed using the multimodel direct inversion algorithm (MMDI) for 2D MRE and a 3D direct inversion algorithm (DI) for 3D EPI MRE to generate quantitative images of tissue stiffness (elastograms).^17,24 Briefly, the 2D MMDI algorithm that is routinely used to calculate tissue stiffness for 2D clinical liver MRE analysis processes each slice independently using four 2D directional filters in the ± X and ± Y directions (cutoff frequencies of two and 128 cycles/FOV [field of view]) to reduce artifacts from wave interference, local polynomial fits to the wave data in 11 × 11-pixel windows to smooth the data, and a Helmholtz wave equation model with an extra constant term to account for longitudinal wave propagation. The correlation coefficient of the local polynomial fit is used as a measure of confidence of the wave data and a m ask indicating confidence values greater than 95% is used as an additional guide during the image analysis described below. The 3D MRE data were processed using 3D, vector MRE techniques, including calculating the curl of the vector wave field (to remove longitudinal wave inform ation), using 20 evenly spaced 3D directional filters (cutoff frequencies of 0.01 and 24 cycles/FOV), and a standard Helmholtz equation inversion to calculate the stiffness. A 3D local frequency estimation (LFE) algorithm was used to calculate a confidence metric related to the local spectral bandwidth of the wave images and a mask indicating confidence values greater than 75% was used as an additional guide during image analysis.

IMAGE ANALYSIS.

The quality of the five MRE acquisitions was assessed by consensus by one radiologist (J.W, 23 years of experience) and one engineer (K.J.G., 18 years of experience) who reviewed the MRE images from each sequence and determined if the MRE analysis was a “success” or “failure” and attributed a reason for any failures without knowing the R2* results. The image quality of the five MRE techniques was evaluated by the readers based on any apparent technique failures, presence and location of artifacts, and the amount of hepatic tissue characterized as reliable for stiffness evaluation by the inversion algorithm. Image quality scores and criteria were defined as follows: 0 = “Unacceptable/Failed”: no regions with trustworthy wave propagation could be observed (eg, due to low SNR or motion artifacts); 1 = “Acceptable”: with or without minor artifacts (eg, artifacts not only at the edge of the liver, typically related to cardiac or respiratory motion); 2 = “Excellent”: free of artifacts or has easily avoided artifacts (eg, artifacts only at the edges, boundaries, or fissures of the liver).

For each patient, the 2D MRE acquisitions were all performed using the same four-slice slice locations (ie, the slice locations for the first scan were copied and pasted into the other 2D MRE scan prescriptions), and those slice locations were used to select four location-matched slices from the 3D MRE data in order to compare the 3D liver results to the other sequences. A manual method was used to select regions of interest (ROIs) for the liver stiffness measurement, and stiffness values were compared among the five methods. Due to the large number of cases that required analysis, four analyzers (J.W., K.J.G., M.Y., S.A.K.; 3–18 years of MRE experience) were used to draw the ROIs. The ROIs were drawn in the right lobe of the liver in regions with observable wave propagation using the MRE magnitude, wave, stiffness, and confidence images as guides. They were drawn to avoid major vessels, focal liver lesions, liver boundaries, and regions with significant wave interference or artifacts. Any questions about the definition of an ROI were resolved by consensus to establish the final ROI from which the mean liver stiffness, in kPa, and the ROI area were recorded.

HISTOPATHOLOGICAL ANALYSIS.

A validation study was performed using 34/332 patients with histological confirmation of liver fibrosis (FO-4, META VIR system) to compare the diagnostic accuracies of the five MRE sequences for staging fibrosis.

Hepatic Fat and Iron Quantification

IDEAL-IQ acquisitions were performed prior to contrast administration.^25,26 The imaging parameters were: TR = 6.2 msec, TE = 2.9 msec, flip angle = 3°, bandwidth = 125 kHz, FOV = 36 cm, matrix size = 160 × 160, slice thickness = 8 mm, and a single 3D slab with 24 slices obtained in one BH of 24 seconds. For R2* measurements in all 332 patients, three ROIs with sizes of 300–400 mm² were placed on the R2* and fat-fraction maps and were manually drawn by one radiologist (3 years of experience) in one IDEAL-IQ liver image. Visible blood vessels, bile ducts, and lesions were avoided in the placement of ROIs. The R2* and fat fraction values from the three ROIs were averaged and the mean values were recorded as the measure of LIC and hepatic fat concentration, respectively, for the patient.

Statistical Analysis

Subject characteristics were summarized using mean ± standard deviation, median (range), and frequencies or percentages as appropriate. Nonparametric comparisons using the Dunn method for joint ranking was used to evaluate differences of the mean values between age, gender, or BMI and R2* or fat fraction. Linear regression was conducted to examine the correlation between the liver stiffness measurements from the five MRE sequences. For each linear regression the intercept and slope of the regression line, the R² coefficient, and root-mean-squared (RMS) error were computed. Bland-Altman plots were generated and used to identify any systematic differences in the stiffness using pairs of the five MRE sequences. The mean of the differences from the Bland-Altman analysis and the 95% confidence interval (Cl) of the mean difference were computed for each comparison. The Wilcoxon signed-rank test was used to compare the measurable parenchyma area (ie, ROI sizes) for the five sequences. Areas under the receiver operating characteristic curve (AUCs) and accuracies for diagnosing early-stage fibrosis (F0–2) and advanced fibrosis (F3–4) were compared among these five techniques. Analysis of covariance (ANCOVA) with standard least squares was used to analyze the effects of BMI, R2*, and fat fraction on the image quality, liver stiffness, and ROI area. All P-value calculations were two-sided, and P < 0.05 was used as the criterion to indicate a statistically significant difference for all statistical tests. All statistical analyses were performed with JMP Pro 11 (SAS Institute, Cary, NC).

RESULTS

Patient Demographics

Upon initial review of the MRE images, 29/361 patients were considered to have had failed MRE exams due to a loose (n = 23) or disconnected (n = 6) MRE driver, resulting in no motion in the liver for analysis, and were excluded from the study. The final study group thus consisted of 332 patients. All five MRE sequences were obtained in 136/332 (41.0%) patients; 196/332 (59.0%) patients were imaged with only the four SE methods.

Figure 1 contains a flowchart showing the subgroups of patients used to assess the success rate and image quality in this study. For the final study group of 332 patients, the average age was 47.9 ± 12.0 years (range: 10–85 years) and the mean BMI was 22.9 ± 3.26 kg/m² (range: 15.4–35.9 kg/m²). The average age for the female (n = 79) and male (n = 253) patients was 46.8 ± 13.6 (F) and 48.2 ±11.5 (M) years. The average BMIs were 22.1 ± 3.08 (F) and 23.1 ± 3.28 (M) kg/m² (range: 16.0–34.2 (F) and 15.4–35.9 (M) kg/m²). There was a significant difference in the mean BMI between males and females (P = 0.022), but no significant difference in mean age (P = 0.374). Chronic hepatitis B (HBV) infection was the primary liver disease prompting the referral for MRE examination (62.7%, 208/332). Thirty-four patients had surgical resection or ultrasound-guided percutaneous liver biopsy at our institution revealing pathologically proven fibrosis (n = 6), hepato-cellular carcinoma (HCC) in = 16), focal nodular hyperplasia (FNH) (n = 2), cholangiocarcinoma (n = 2), metastatic tumor (n = 3), hemangioma (n = 1), hepatic adenoma (n = 2), Budd-Chiari syndrome (n = 1), and rejection reaction after liver transplantation (n = 1).

FIGURE 1:

Flowchart for our retrospective study. The study cohort was divided into several subgroups to evaluate the success rate and image quality of the standard 2D GRE MRE sequence and four SE MRE sequences.

Distribution of R2*

The mean fat fraction for all 332 patients was 4.16% (median: 3.10%, range: 0.37–34.9%). Table 3 shows that there was a broad left-skewed distribution of the R2* values for these patients with a mean value of 67.9 s^–1 (median:51.2 s^–1, range: 19.5–341 s^–1). There was a weak correlation between R2* and fat fraction (R2 = 0.0242, P < 0.0001) and between fat fraction and BMI (R2 = 0.155, P < 0.0001). The R2* was significantly greater in males (72.2 s^–1) than in females (53.3 s^–1 P < 0.001). Nonpara-metric comparisons using the Dunn method for joint ranking test showed a weak correlation between R2* and gender (r = 0.279, P < 0.001). There was no significant difference in the mean BMI between patients with R2* ≥ 100 s⁻¹ (elevated R2*) and R2*< 100s⁻¹ (P, = 0.05). The prevalence of patients with elevated R2* was 9.6% (32/332).

TABLE 3.

R2*-Dependent Success Rates and Image Quality Assessments of the 5 MRE Sequences

	R2* Range	# of Patients	3D EPI MRE	Iron 2D EPI MRE	Iron 2D SEMRE	Non-iron 2D EPI MRE	2D GRE
	<70 s−1	240(72.3%)	240/240(100%)	239/240(99.6%)	240/240(100%)	240/240(100%)	100/100(100%);
R2* Value Range	70–100 s−1	60(18.1%)	60/60(100%)	60/60 (100%)	60/60(100%)	60/60(100%)	16/27(59.3%)
	100–200 s−1	24(7.23%)	23/24(95.8%)	24/24(100%)	24/24(100%)	23/24(95.8%)	0/8(0%)
	200–400 s−1	8(2.41%)	6/8(75.0%)	8/8(100%)	8/8(100%)	7/8(87.5%)	0/1(0%)
	>400 s−1	0	0/0	0/0	0/0	0/0	0/0
	Total	332	329/332(99.1%)	331/332(99.7%)	332/332(100%)	330/332(99.4%)	116/136(85.3%)
Image Quality	Grade 0		1(0.74%)	0	0	0	20(14.7%)
	Grade 1		0	18(13.2%)	18(13.2%)	3(2.21%)	56(41.2%)
	Grade 2		135(99.3%)	118(86.8%)	118(86.8%)	133(97.8%)	60(44.1%)
		136	136	136	136	136	136

2D: two-dimensional; 3D: three-dimensional; EPI: echo-planar imaging; GRE: gradient-recalled echo; MRE: magnetic resonance elastography; SE: spin echo.

MRE Success Rate and Image Quality Comparisons

MRE SUCCESS RATE.

The R2* distribution of all 332 patients and the success rates of the five MRE sequences are summarized in Table 3. The success rates are significandy higher for the four SE sequences than for GRE MRE (all P < 0.001). Two hundred and forty patients (240/332,72.3%) had R 2*< 70 s^–1, with nearly 100% of these patients having successful MRE exams using all five sequences (Fig. 2). The two iron 2D SE/EPI MRE sequences were successful in all patients who had R2* ≥ 100 s^–1, but failed in one patient with R2 *< 100 s⁻¹. The noniron 2D EPI MRE sequence succeeded in all patients, with R2* < 100 s^–1 but failed in two patients with R2* ≥ 100 s^–1 (Fig. 3).

FIGURE 2:

A 48-year-old male patient with left lateral lobectomy, (a) The liver parenchyma was normal in the T2-weighted images (T₂WI). (b,c) The R2* and fat fraction were 48 s⁻¹ and 2.16%. The five MRE sequences were successful, with the largest ROI and artifact-free region obtained using 3D EPI MRE (d) compared with the other 2D MRE sequences (e-h). The checkerboard mask overlaying the elastograms indicates regions the MRE inversion processing considered to have low confidence based on the image SNR and wave quality and was used as a guide for drawing ROIs from which to report the liver stiffness.

FIGURE 3:

A 52-year-old male patient with cirrhosis and portal hypertension due to HBV infection. The liver parenchyma was dark in the T₂WI (a) with a high R2* of 320 s⁻¹ (b). The 3D EPI and noniron 2D EPI sequences failed due to low SNR (c,d) whereas the iron 2D SE/EPI sequences were successful (e,f).

The 3D EPI MRE sequence succeeded in all patients with R2* < 100 s^–1 but, failed in three patients with R2* ≥ 100 s^–1 (Fig. 3). Among the 136 patients who underwent the 2D GRE MRE exam, GRE MRE failed in all nine cases (100%, 9/9), with R2* > 100 s^–1 and in 11 cases (40.7%, 11/27) with R2* between 70 and 100 s⁻¹ (Figs. 4 and 5).

FIGURE 4:

A 66-year-old male patient with HBV infection. The liver parenchyma was dark in the T₂WI (a) with a high R2* of 215 s-1 and a fat fraction of 3.81% (b,c). All four of the SE MRE sequences were successful (d-g), but 2D GRE MRE failed due to low SNR (h).

FIGURE 5:

A 48-year-old male patient with nodular cirrhosis and portal hypertension due to HBV infection. The liver parenchyma was dark in the T₂WI (a) with an R2* of 150 s⁻¹ and a fat fraction of 2.91% (b,c). All four SE MRE sequences were successful (d-g), but 2D GRE MRE failed due to low SNR (h).

MRE IMAGE QUALITY.

For the 136 patients with all five MRE sequences available, the MRE image quality was graded on a scale from 0 (“unacceptable/failed”) to 2 (“excellent”) as summarized in Table 3.

Liver Stiffness and ROI Area

For the 136 patients with all five MRE sequences available, 115 patients (84.6%) had successful MRE exams (grade-1 or grade-2 image quality) for all five sequences (Table 3). The 21 unsuccessful cases included 20 patients with GRE MRE failures and one patient with a 3D MRE failure. The mean liver stiffness was measured and compared using the five MRE sequences for these 115 patients (Table 4). The mean stiffness for these 115 patients was 4.07 ± 2.27 kPa, 3.81 ± 2.3 8 kPa, 4.22 ± 2.38 kPa, 4.11 ± 2.20 kPa, and 4.21 ± 2.28 kPa using 2D GRE, 3D EPI, noniron 2D EPI, iron 2D EPI, and iron 2D SE, respectively. The Wilcoxon signed-rank test showed that there was a statistically significant difference in the stiffness values measured using 3D EPI and any 2D SE sequence (all P < 0.0001), iron 2D EPI and 2D GRE (P = 0.0348), and iron 2D EPI and iron 2D SE (P = 0.0061). Linear regression analysis showed a high correlation of the stiffness measurements from any pair of MRE sequences, with R2 ranging from 0.9053–0.9714 (P < 0.005). The Bland-Altman analysis showed the bias and 95% confidence limits of the mean difference between each pair of sequences. The bias ranged from −0.2313 to 0.3809 kPa, which is less than 10% of the mean liver stiffness for this cohort of patients.

TABLE 4.

Pairwise Regression and Bland-Altman Analysis of the Liver Stiffness Measured Using the 5 MRE Sequences

	Linear regression				Bland-Altman
	Intercept	Slope	R2	Root mean squared error	Bias	95% confidence interval
2D GRE vs. 3D EPI	0.6956	0.8792	0.9053	0.7023	−0.2313	[−0.3722, −0.0903]
2D GRE vs. Noniron 2D EPI	0.2386	0.9083	0.9564	0.4772	0.1497	[0.0526, 0.2468]
2D GRE vs. Iron 2D EPI	0.0571	0.9729	0.9334	0.5862	0.0520	[−0.0568, 0.1607]
2D GRE vs. Iron 2D SE	0.1058	0.9398	0.9272	0.6131	0.1499	[0.0337, 0.2661]
3D EPI vs. Non-iron 2D EPI	−0.2412	0.9604	0.9243	0.6577	0.3809	[0.2572, 0.5047]
3D EPI vs. Iron 2D EPI	−0.4559	1.0339	0.9153	0.6945	0.2832	[0.1529, 0.4136]
3D EPI vs. Iron 2D SE	−0.4078	0.9986	0.9162	0.6911	0.0646	[0.2532, 0.5091]
Non-iron 2D EPI vs. Iron 2D EPI	−0.1671	1.0636	0.9687	0.4233	−0.0977	[−0.1772, −0.0183]
Non-iron 2D EPI vs. Iron 2D SE	−0.1361	1.0304	0.9714	0.4041	0.0002	[−0.076, 0.0764]
Iron 2D EPI vs. Iron 2D SE	0.1080	0.9509	0.9685	0.3916	0.0979	[0.0277, 0.1681]

2D: two-dimensional; 3D: three-dimensional; EPI: echo-planar imaging; GRE: gradient-recalled echo; MRE: magnetic resonance elastography; SE: spin echo; R: correlation coefficient.

The fibrosis staging capability of the five MRE sequences was assessed using 34 patients (31 male [92%], three female [8%]), with ages ranging from 10–78 years (mean ± SD, 44 ± 1 1 years). The liver stiffness values obtained for each fibrosis stage and MRE sequence are shown in Table 5. There were no significant differences in staging advanced fibrosis between 2D GRE and 3D EPI (P = 0.4795), 2D GRE and noniron 2D EPI (P = 0.4795), 2D GRE and iron 2D EPI (P = 0.4042), 2D GRE and iron 2D SE (P = 0.4795), 3D EPI and noniron 2D EPI (P = 0.2410), 3D EPI and iron 2D EPI (P = 0.3548), 3D EPI and iron 2D SE (P = 0.7611), noniron 2D EPI and iron 2D EPI (P = 0.9171), noniron 2D EPI and iron 2D SE (P = 0.4127), and iron 2D EPI and iron 2D SE (P = 0.3745). The AUCs for discriminating advanced fibrosis (F3–4) by 2D GRE, 3D EPI, noniron 2D EPI, iron 2D EPI, and iron 2D SE were 0.917 (95% Cl: 0.742–1.091), 0.944 (Cl: 0.814–1.075), 0.944 (Cl: 0.814–1.075), 0.972 (Cl: 0.890–1.055), and 0.889 (Cl: 0.672–1.106), respectively (all PC 0.05). There were no significant differences of the AUCs of the five MRE sequences for discriminating advanced fibrosis (P > 0.05).

TABLE 5.

Mean Liver Stiffness (kPa) for the Patients in Each Fibrosis Stage as Measured by the 5 MRE Sequences

	Liver stiffness (Mean ± SD)
Fibrosis stage	2D GRE (n = 13/332)	3D EPI MRE (n = 34/332)	Non-Iron 2D EPI MRE (n = 34/332)	Iron 2D EPI MRE (n = 34/332)	Iron 2D SE MRE (n = 34/332)
F0	2.38 ± 0.45(1)	1.92 ± 0.24(2)	2.03 ± 0.56(2)	1.99 ± 0.44(2)	2.09 ± 0.55(2)
F1	3.09 ± 0.29(2)	2.08 ±0.16(3)	2.46 ± 0.35(3)	2.72 ± 0.32(3)	3.02 ± 0.36(3)
F2	2.11 ±0.37(1)	2.97 ±0.35(5)	3.67 ± 0.98(5)	3.82 ± 0.80(5)	3.28 ± 0.56(5)
F3	3.59 ± 0.54(3)	2.83 ± 0.260(6)	3.44 ±0.71(6)	3.29 ±0.51(6)	3.29 ± 0.49(6)
F4	4.62 ± 0.64(6)	4.55 ± 0.64(18)	5.12 ±0.73(18)	5.15 ±0.94(18)	5.23 ±0.88(18)

2D: two-dimensional; 3D: three-dimensional; EPI: echo-planar imaging; GRE: gradient-recalled echo; MRE: magnetic resonance elastography; SE: spin echo; SD: standard deviation.

The mean ROI area (size) for the MRE stiffness measurement from the 115 patients with all five successful MRE exams was 667, 1281, 1151, 799, and 359 mm² using 2D GRE, 3D EPI, noniron 2D EPI, iron 2D EPI, and iron 2D SE, respectively. The Wilcoxon signed-rank test showed that the statistically significant largest and smallest ROI areas were from 3D EPI (P < 0.0001 for all) and iron 2D SE (P < 0.0001 for all), respectively. There were statistically significant differences of the mean ROI area between every pair of MRE sequences (P-values from 0.0014 to <0.0001).

Effects of BMI, R2*, and Fat Fraction on MRE Quality, Liver Stiffness, and ROI Area

The ANCOVA test showed that the R2* value had a significant Effect on image quality for noniron 2D EPI (P = 0.0019), 3D EPI {P < 0.0001), and 2D GRE (P < 0.0001), but no significant Effect on iron 2D EPI (.P = 0.1203) and iron 2D SE (P = 0.1156). BMI had a significant Effect on image quality for iron 2D EPI (P = 0.0230) and iron 2D SE (P = 0.0040). Fat fraction was not associated with the image quality of the five MRE sequences. BMI, R2*, and fat fraction had no significant Effect on the measured liver stiffness (P-value: 0.2487–0.6094) or the ROI area (P-value: 0.2903–0.9529) for the five sequences.

DISCUSSION

The results from this preliminary, retrospective study support the hypothesis that SE sequences have better image quality and a higher success rate for routine clinical 3T hepatic MRE than the conventional 2D GRE sequence. In previous studies,^{16,18,23,27,28} various SE MRE sequences were shown to have shorter acquisition times and better image quality for normal subjects and patients with and without iron overload. The results of this study provide a more comprehensive comparison of four implementations of SE MRE and GRE MRE with additional information about dependencies on hepatic R2* and fat fraction.

The prevalence of patients with iron overload (R2* > 100 s^–1) in this study was low, as was the failure rate due to low SNR for the “standard-TE” SE sequences. The short-TE, iron-overload SE sequences also had higher success rates and better image quality than 2D GRE. There was a higher failure rate of 2D GRE MRE in these patients compared with an earlier clinical study by Yin et al, but similar to a recent clinical study by Wagner et al.^8,29 This could be caused by the different magnetic field strengths (1.5T vs. 3.0T) and the different patient demographics. In this study, as in the work by Yin et al, BMI was not associated with GRE MRE failure, whereas an association was found in Wagner et al. It is not clear why there is a difference between the three studies. The results from this study also agree with Wagner et al that hepatic fat fraction is not associated with GRE MRE failures.

2D GRE MRE images often have artifacts around the edges of the liver due to susceptibility effects; effects which get worse at higher field strengths and are avoided using SE imaging. The EPI acquisition with 3D vector MRE processing may also reduce the artifacts that arise due to through-plane wave propagation and wave scattering or diffraction effects that cause problems in 2D MRE analysis.^17,30,31 The noniron 2D EPI sequence achieved the highest image quality and highest success rate among the four SE sequences, typically only suffering from a possible cardiac-induced motion artifact in the left lobe, whereas the iron 2D SE/EPI sequences sometimes had ghosting artifacts, possibly due to the short TR and reduced flow compensation, causing some problems maintaining a steady-state MR signal.

All five MRE techniques show good diagnostic performance for discriminating early-stage fibrosis (FO-2) and advanced fibrosis (F3–4), and the diagnostic performances were statistically equivalent. This indicates that any differences in the stiffness values among the five MRE sequences may not be important when using them to diagnose liver fibrosis. This result is in agreement with recent literature.^32,33 The different ROI sizes used in the five MRE sequences may have an effect on the stability of the stiffness measurements because of the liver parenchyma stiffness heterogeneity often observed in MRE elastograms. While previous work has established the validity of 3D MRE when compared to mechanical testing of phantoms,³⁴ it has also been observed that 3D hepatic MRE stiffness values can be lower than 2D MRE values, while having comparable diagnostic accuracy and fewer failures.^32,35

The 3D MRE stiffness measurements may be lower than the 2D measurements for several reasons. First, through-plane propagating waves can appear in 2D data as waves with longer wavelengths, thereby resulting in an overestimate of the tissue stiffness. Second, the thin-slice acquisition used in this work to produce a 3D sampling of the wave field can have lower SNR than the standard thick-slice 2D MRE acquisitions, thus potentially causing an underestimate of the stiffness. Third, The 3D vector MRE data were processed using a different inversion algorithm than what was used for 2D MRE, which may result in a relative difference in the stiffness and confidence estimates in some cases. Finally, since 3D MRE had larger ROI areas, the heterogeneity within the elastograms and the larger ROI areas for these patients may have introduced an additional bias (eg, by including more tissue deeper in the liver that may appear softer than more superficial hepatic tissue).

In our study, BMI, R2*, and fat fraction had no significant Effect on the liver stiffness measurements for the five sequences. R2* had no significant Effect on MRE image quality and BMI did have a significant Effect on the quality of the iron 2D SE/EPI sequences. This is because the SE sequences are more dependent on hepatic T2 rather than T*₂ decay, and the short TE of the iron-overload sequences further reduces the signal decay. These results show that the SE techniques may be accurate measures of liver stiffness that are comparable to the conventional GRE technique and that these SE techniques may be reliable for the diagnosis of cirrhosis in patients with diffuse liver disease and iron overload. The effect of steatosis on the MRE measurement of liver stiffness has varied in different studies. It has been reported that fat fraction does not significantly correlate with liver stiffness at Each fibrosis stage and thus does not appear to interfere with the ability of MRE to assess fibrosis extent.^36,37 However, Yoon et al found that livers with steatosis had a higher stiffness than normal livers. This discrepancy could be explained by the different study populations and study profile.³⁸

There were several limitations in this study. First, histological proof about iron content, fibrosis, and cirrhosis was not available for most of these patients. Only 34/332 cases had biopsy-proven fibrosis and cirrhosis, which is too small a sample to assess the diagnostic performance of the five sequences. However, this limitation is common, as it is generally not common practice to obtain biopsies for every patient in routine clinical practice.³⁹ Second, prospective studies are required for evaluating the utility of these sequences for assessing and monitoring chronic diffuse liver disease. Third, due to the relatively poor quality of the GRE MRE images compared to the SE images, the clinical protocol was changed to not include the GRE MRE acquisition to save time. This reduced the number of patients for whom data from all five sequences were available. As a result, only a fraction of the cohort was used for the overall comparison. Fourth, 21 cases were excluded due to technical failures unrelated to the factors studied here (eg, the passive and active driver components were not connected, so no motion was produced in the liver for analysis), which further reduced the number of patients available for study. Fifth, measurement reproducibility and repeatability were not studied. Sixth, the semiquantitative and subjective image quality grading system used in this study could be improved. Future work will include validation of the hepatic fibrosis measurements performed in iron-overloaded livers using these techniques at 3.0T.

The stiffness measurements for these five MRE sequences were highly correlated with each other. Compared to 2D GRE MRE, the SE MRE sequences have a higher success rate (>99%) and better image quality at 3.0T in a population with a spectrum of LIC. All five techniques showed good diagnostic performance in staging liver fibrosis, and the diagnostic performances were statistically equivalent. R2* had a significant Effect on the success rate and image quality in the noniron 2D EPI, 3D EPI, and 2D GRE sequences. While 3D EPI MRE had the best image quality and the largest reliable ROI area, BMI, R2*, and fat fraction had no significant Effect on the liver stiffness measurement or the ROI area for any of the five sequences.