Quality of Gadoxetate-enhanced MRI versus US during Hepatocellular Carcinoma Screening in Participants with Cirrhosis.

Introduction
Hepatocellular carcinoma (HCC) is a major global health challenge standing as the
third leading cause of cancer-related death, also representing a significant public
health concern in the United States, with more than 30 000 new cases annually
and a 5-year survival rate under 20%, due to its aggressive tumor biology and the
competing risk from comorbid chronic liver disease (1,2). Early detection enables
curative treatment and improves outcomes, which has prompted guidelines recommending
surveillance in patients who are at risk of HCC by the American Association for the
Study of Liver Diseases, or AASLD, and the European Association for the Study of the
Liver, or EASL (3,4).
A landmark Chinese randomized controlled trial on surveillance in mostly patients
with hepatitis B without cirrhosis showed that semiannual US combined with
α-fetoprotein testing reduced HCC-related mortality by 37% compared with no
surveillance (5). These results are supported
by recent meta-analyses of other cohort studies demonstrating that HCC surveillance
is associated with improved early-stage, curative treatment receipt and overall
survival (6). HCC surveillance every 6 months
using US and α-fetoprotein testing is the recommended practice standard in
the United States (3), especially given the
widespread availability and accessibility of US. However, there is no level 1
evidence in patients with cirrhosis (7).
Importantly, the results of the Chinese surveillance study cannot be extrapolated to
Western populations, which are impacted by a high prevalence of obesity and
cirrhosis, factors that reduce the sensitivity of US for early HCC detection (8). A previous meta-analysis demonstrated a 47%
sensitivity of US for early HCC in cirrhosis (8). The diagnostic performance of US also suffers from interoperator
variability (9) and poor liver visualization
scores in many cases (9,10), as well as poor compliance (11). The limitations of the current surveillance paradigms are
manifold, resulting in the failed detection of many early HCCs, contributing to
delayed diagnosis and increased patient mortality (12).
MRI is increasingly used for HCC surveillance given its higher sensitivity, even if
US remains the most frequently used surveillance modality across most U.S. centers
(13,14). A previous study from a large tertiary care North American health
system showed that the index HCC surveillance imaging modalities were MRI (51%), US
(33%), and CT (16%) (15). Gadoxetate-enhanced
(Eovist, Primovist; Bayer Healthcare) MRI may further improve sensitivity for HCC
surveillance (16,17), with potential cost effectiveness and improved efficiency
in certain scenarios (18). A challenge in the
use of gadoxetate is the risk of motion artifacts at arterial phase (AP) imaging
(19). Nonetheless, a prospective study
from South Korea showed superior sensitivity of gadoxetate-enhanced MRI compared
with US for early-stage HCC detection (86% vs 29%) (20).
Identifying patient or technical factors that affect image quality, referring to the
technical performance of an imaging system (21), is key to optimizing HCC surveillance. However, comparative data
for MRI and US in the same patients are lacking.
The purpose of this study is to compare image quality and its determinants in
gadoxetate-enhanced MRI versus US in participants with cirrhosis undergoing HCC
screening in a prospective bicenter study.

Materials and Methods

Study Sample
This is a Health Insurance Portability and Accountability Act–compliant,
prospective, nonrandomized, single-arm North American multicenter trial
(involving sites in the United States) with a centralized institutional review
board approval. Signed informed consent was obtained from all participants. This
study was performed from September 2020 to May 2023 in accordance with the
Declaration of Helsinki.
The main objectives of the broader trial are to assess the diagnostic performance
of hepatobiliary phase (HBP)–abbreviated MRI versus US plus
α-fetoprotein testing for HCC screening, to assess the added value of
blood cell free tumor DNA for improving HCC detection; and to compare the
cost-effectiveness of HBP-abbreviated MRI versus US plus α-fetoprotein
testing for HCC screening in participants with cirrhosis. The full study
protocol (Focused Abbreviated Screening Technique [FAST]–MRI Study,
NCT04539717) is available at site 1 (Icahn School of Medicine at Mount Sinai,
the coordinating site). This current study includes data from two of five study
sites (sites 1 and 2 [Duke University Medical Center]).
Inclusion criteria for participants were (a) age older than 18
years, (b) liver cirrhosis of any etiology,
(c) enrolled in a surveillance program for HCC,
(d) scheduled for standard-of-care surveillance MRI,
(e) estimated glomerular filtration rate of more than 30
mL/min/1.73 m2, (f) no negative surveillance imaging
for HCC less than 5 months before enrollment, (g) no prior
liver resection or transplant, (h) no history of Liver Imaging
Reporting and Data System (LI-RADS) category 3 or 4 lesion, and
(i) no prior treatment for HCC. Exclusion criteria included
(a) patients who withdrew consent to participate after
enrolling or (b) patients who were unable to complete both MRI
and US (Fig 1). Cirrhosis was established
by the referring hepatologist using blood tests (Fibrosis-4 index), transient
elastography, histopathology, and/or imaging evidence of liver surface
nodularity.
Participants underwent clinical gadoxetate-enhanced MRI and research US.
Demographic and clinical data were extracted from the medical record, including
age, sex, self-reported race, self-reported ethnicity, body mass index, and
liver disease etiology. Race and ethnicity were recorded as part of the study
protocol. Calculated clinical and/or laboratory scores included Child-Pugh
score, albumin-bilirubin grade, Model for End-Stage Liver Disease (known as
MELD) score, and sodium-MELD score. All data were anonymized and centralized in
site 1.

MRI Acquisition
Standard-of-care liver MRI examinations were performed at sites 1 and 2 using
clinical MRI systems, including 1.5-T systems (MAGNETOM Aera, Avanto, and Amira
[Siemens Healthineers]; Signa [GE HealthCare]; n = 79) and 3-T
systems (Magnetom Skyra [Siemens Healthineers]; Signa Premier [GE HealthCare];
n = 166). Noncontrast and dynamic contrast-enhanced imaging
used breath-hold three-dimensional fat-suppressed spoiled gradient-echo
T1-weighted sequences (volumetric interpolated breath-hold examination [VIBE] or
liver acquisition with volume acceleration [LAVA]). Images were acquired before
contrast material administration, in early and late APs (dual arterial phases at
site 1; triple arterial phases at site 2), portal venous phase (60 seconds),
transitional phase (180 seconds), and HBP (10 and 20 minutes after injection).
Siemens systems used bolus tracking; GE systems used a fixed 20-second delay.
Gadoxetate disodium (0.025 mmol/kg, with a maximum dose of 10 mL) was injected
at 2 mL/sec with a 20-mL saline flush. Free-breathing single-shot spin-echo
echo-planar diffusion-weighted imaging (DWI) was performed after contrast
material administration (b values, 50 and 800
sec/mm2). Apparent diffusion coefficient maps were automatically
generated online on a pixel-by-pixel basis assuming monoexponential decay.
T2-weighted (half-Fourier acquisition single-shot turbo spin-echo imaging or
single-shot fast spin-echo imaging) imaging and T1-weighted in- and
opposed-phase gradient-echo sequences were also performed as part of the
protocol.
Examination length was recorded. All studies were prospectively interpreted for
clinical care and reviewed retrospectively for this analysis.

US Acquisition
Abdominal US examinations were performed at outpatient imaging facilities at the
two sites contemporaneous to the MRI examination (Philips IU-22, Epiq 7, Epiq 5,
and Epiq Elite; Philips Healthcare). All US examinations were performed on the
same day as MRI at site 1. One hundred five US examinations were performed on
the same day as MRI at site 2, whereas the remaining 14 US examinations were
performed within a mean delay of 6.2 days (range, 1–14 days) of MRI. The
US examinations were performed by registered diagnostic medical sonographers at
each site during the study period according to a standardized protocol in
accordance with LI-RADS recommendations (22,23). Examination length
was recorded.

MRI Quality Scoring
Two independent subspecialty-trained abdominal radiologists (observer 1, S.L.,
site 1 with 13 years of experience; observer 2, B.W.T., at site 2 with 6 years
of experience) reviewed 126 and 119 MRI examinations from their respective
sites. Both observers participated in the original clinical MRI interpretations.
The observers were blinded to the US findings.
The quality of the images from dynamic axial T1-weighted sequences (before
contrast material administration, AP, portal venous phase) and HBP (at 20
minutes) was graded using a five-point score, as previously reported (24), as follows: 1, no motion; 2, minimal
motion, without effect on diagnostic quality; 3, moderate motion, with minimal
effect on diagnostic quality; 4, severe motion, with limited diagnostic quality;
and 5, severe motion, nondiagnostic. The AP sequence with the least motion and
optimal timing (early portal vein enhancement without hepatic vein enhancement)
was scored. Scores across sequences were summed (range, 4–20) for a
“dynamic MRI motion score.” Transient severe motion was assessed,
as defined by an AP score of at least 4 with a score of 3 or less in other
dynamic phases (24). HBP contrast
material uptake at 20 minutes was assessed on axial T1-weighted images, graded
on a three-point scale as published previously (25): 1, no or minor limitations (liver parenchyma hyperintense
compared with the portal vein, indicating adequate gadoxetate uptake); 2,
moderate limitations (liver parenchyma isointense to portal vein); and 3, severe
limitations (liver parenchyma hypointense to portal vein, indicating severely
impaired gadoxetate update). Limited HBP liver contrast material uptake was
defined as a score of at least 2.
DWI was assessed for anatomic distortion, sharpness, echo planar imaging
artifacts, motion, and apparent diffusion coefficient map quality using a
five-point scale (26), as follows: 1,
excellent, 2, good, 3, satisfactory, 4, poor, and 5, unacceptable. Summed DWI
scores ranged from 5 to 25. Dynamic motion and DWI scores were dichotomized as
diagnostic (scores 1–3) and nondiagnostic (scores 4–5).
A global MRI score was computed by summing motion (scores 4–20), HBP
uptake (scores 1–3), DWI quality (scores 5–25), for a total score
range of 10–48, and classified as follows: MR-A (scores 10–22),
meaning no or minor limitations; MR-B (scores 23–35), meaning moderate
limitations; and MR-C (scores 36–48), meaning severe limitations. MR-A
and MR-B were considered diagnostic, and MR-C was considered nondiagnostic. The
MRI quality scoring system (Table S1) resembles the Prostate Imaging Quality, or PI-QUAL,
scoring system, using semiquantitative 1–5 scales without sequence
weighting (27). The observers recorded
ascites semiquantitatively (none, mild, moderate, severe) (19). The clinical MRI reports for MR-B and MR-C were
reviewed by observer 1 for terms indicating reduced quality (eg,
“limited,” “degraded,”
“nondiagnostic”).
Before image analysis, the observers underwent a calibration training session on
15 reference cases from a separate published study, scores from which were
considered the ground truth (19). The
purpose of the training session was to harmonize the observers’
assessment of image quality. Twenty separate MRI examinations (“test
set”), also not included in the current study database, were reviewed by
both observers for assessment of interobserver reproducibility. The observers
were blinded to the participants’ clinical, demographic, pathologic, and
laboratory data.

US Quality Scoring
Although US was performed for research purposes, the US examinations were
interpreted independently by board-certified, fellowship-trained abdominal
radiologists as part of routine clinical practice, with expertise in US locally
at each center with a range of experience from 2 to 25 years, using a picture
archiving and communication system (site 1: Centricity 3.0 [GE Medical Systems];
site 2: Visage [Visage Imaging]). The US interpreters were blinded to the MRI
findings. US liver visualization scores based on the LI-RADS Surveillance US
algorithm were extracted from the standardized clinical interpretations. The US
LI-RADS quality score includes US-A (studies in which the liver is homogeneous
or minimally heterogeneous, there is minimal beam attenuation or shadowing, or
where the liver is visualized in near entirety), US-B (studies in which the
liver is moderately heterogeneous, there is moderate beam attenuation or
shadowing, or some portions of the liver or diaphragm are not visualized), and
US-C (studies in which the liver is severely heterogeneous, there is severe beam
attenuation or shadowing, or the majority [>50%] of the liver and/or
diaphragm are not visualized) (28). These
criteria have been validated in large studies (29,30). For the purposes of
this study, US-A and/or US-B were considered diagnostic and US-C was considered
nondiagnostic.

Statistical Analysis
This analysis reflects a per-protocol assessment among participants eligible for
MRI-based surveillance, as subject inclusion in this comparative study focusing
on image quality requires successful completion of both modalities. Demographic,
clinical, and imaging data were summarized and dichotomized according to US
liver visualization and MRI scan quality (score A vs B and C). Differences in
baseline characteristics between the participants at site 1 and site 2 were
assessed using the Wilcoxon rank sum and Fisher exact tests. Covariates analyzed
for quality association included age, sex, race and ethnicity, body mass index
(BMI; obesity defined as BMI ≥ 30, calculated as weight in kilograms
divided by height in meters squared), and liver function measures (Child-Pugh,
MELD, and sodium-MELD scores). Univariable associations were tested with
χ2 (categorical variable) and two-sided t
(continuous variables) tests. The strength of association was quantified using
odds ratios (ORs) and 95% CIs for those variables showing significant
association. Multivariable generalized linear mixed models were fit by maximum
likelihood incorporating random effect by site and vendor. The adjusted OR (aOR)
(with 95% CI) for significant covariates was summarized.
MRI quality scores were compared with US liver visualization scores for
concordance. Interobserver reproducibility was calculated for MRI in the test
set of 20 examinations (binarized as 1–3 vs 4–5 for dynamic MRI
motion scores and DWI quality scores; HBP liver uptake scores were binarized as
1 vs 2–3). P values of less than .05 were considered
statistically significant. Analysis was performed in R, version 4.2.3 (The R
Foundation for Statistical Computing; https://www.R-project.org) (H.J.).

Results

Participants
This study included 245 participants (median age, 61 years; IQR, 54–68
years; range, 22–80 years; 133 men) recruited from two centers (site 1,
126 participants; site 2, 119 participants), with a few differences between the
two populations (Table 1). Exclusion
criteria included patients who withdrew consent to participate after enrolling
(site 2, three patients) or those unable to complete both MRI and US (site 2,
three patients) (Fig 1).

Interobserver Agreement
Interobserver concordance ranged from 0.75 to 0.95 for each MRI dynamic motion
and DWI quality score (Table
S2). Kappa values could not be computed as there was 100% agreement
in at least one category, indicating strong agreement between observers (31).

MRI Acquisition and Quality
The mean (±SD) contrast material volume was 8.9 mL ± 1.5, with an
average MRI examination time of 38.3 minutes ± 10.1. Most MRI scans were
diagnostic, with 89.0% (218 of 245) of AP images rated as diagnostic (motion
score, 1–3) and only 1.2% showing transient severe motion. The HBP liver
uptake scores were 1 (no or minor limitations) in 92.7% (227 of 245), 2
(moderate limitations) in 6.5% (16 of 245), and 3 (severe limitations) in 0.8%
(two of 245) of participants. The global MRI quality score was classified as
MR-A in 80.4%, MR-B in 18.4%, and MR-C in 1.2% of participants, with 98.8% of
MRI scans considered diagnostic. Only 10.4% (five of 48) of MR-B and MR-C cases
had clinical reports indicating reduced quality. These results are summarized in
Table 2.

US Imaging
The mean length of US examination was 15.5 minutes ± 8.8. Among 240
participants, liver visualization scores were classified as US-A (no or minimal
limitations) in 58 participants (24.2%), US-B (moderate limitations) in 148
participants (61.7%), and US-C (severe limitations) in 34 participants (14.2%)
(Table 2). In five participants
(2.0%), the radiology report omitted the visualization score.

Relationship between MRI and US Quality Scores
Compared with US, MRI had a higher proportion of scores with no or minimal
limitations in the whole study sample (MR-A, 80.4% [193 of 240 participants] vs
US-A, 24.2% [58 of 240 participants]; McNemar P < .001)
and in the subset of participants with obesity (MR-A, 73.1% [95 of 130
participants] vs US-A, 18.5% [24 of 130 participants]; McNemar
P < .001), for those with both sets of scores (Table 3). However, there was no evidence
of an association between MRI global quality scores and US liver visualization
scores for A compared with B and C for the whole study sample (χ2
P = .27) and for the subset with obesity
(χ2 P > .62), indicating that
there was no interdependence between the quality scores. For the 182
participants with US-B or US-C liver visualization scores, 143 participants
(78.6%) had an MRI quality score of MR-A. Conversely, of the 47 participants
with MRI quality scores of MR-B or MR-C, 39 (83.0%) had US-B or US-C liver
visualization scores.

Predictive Factors of MRI and US Quality
Table 4 shows clinical and demographic
factors associated with reduced MRI and US quality (scores B or C). At
univariable analysis, reduced MRI quality was associated with non-Hispanic
ethnicity (P = .14), obesity (P = .005), and
higher Child-Pugh score (P = .03). Hispanic ethnicity was
associated with lower odds of reduced MRI quality (OR, 0.12 [95% CI: 0.00,
0.77]; P = .01), though the sample was small (12.2%, 30 of 245
participants). Hispanic ethnicity was not significant at multivariable analysis
(multivariable aOR, 1.4 × 106 [95% CI: 0, infinite];
P = .93), which was likely due to model instability from
sparse data. Obesity (OR, 4.20 [95% CI: 1.20, 22.6]; P = .01)
increased odds of reduced MRI quality and US quality (OR, 2.6 [95% CI: 1.00,
6.00]; P = .04) at univariable analysis (Fig 2). However, obesity was not significant at
multivariable analysis for reduced MRI quality (aOR, 2.19 [95% CI: 0.47, 10.11];
P = .32) or US quality (aOR, 1.77 [95% CI: 0.68, 4.61];
P = .25). There was a weak association between being
overweight (BMI, 20–30) and reduced image quality scores (Pearson
correlation between the numeric values of BMI and global MRI score, 0.34 [95%
CI: 0.22, 0.45]; P < .001). Figure 3 illustrates the relationship between BMI and
global MRI quality scores. Child-Pugh B or C (OR, 2.6 [95% CI: 1.05, 6.30];
P = .03) increased the odds of reduced MRI quality at
univariable analysis and was confirmed at multivariable analysis (aOR, 3.95 [95%
CI: 1.50, 10.42]; P = .006; R2 =
0.8). Figure 4 illustrates an example of
a participant with excellent MRI and US quality. Figure 5 illustrates an example of a participant in whom US quality
was better than that of MRI.

Discussion
The lack of prospective randomized data in Western countries validating
hepatocellular carcinoma (HCC) screening in patients with cirrhosis has limited the
knowledge on its effectiveness, especially in an overweight population. Few
prospective studies have compared MRI and US directly in the same patients, and
fewer assessed image quality (20,32). Although the Liver Imaging Reporting and
Data System standardizes liver imaging interpretation, it offers limited guidance on
evaluating image quality, although early efforts are underway (21). Our multicenter HCC screening study, reporting data from
two of five sites, found that a greater proportion of participants had excellent
quality with gadoxetate-enhanced MRI compared with US (MR-A, 80.4% [193 of 240
participants] vs US-A, 24.2% [58 of 240 participants]; P <
.001), even though obesity reduced quality across both modalities (at univariate
analysis). Notably, poor US quality did not predict poor MRI quality
(P = .27). These findings suggest that obese patients with
cirrhosis often have reduced US quality, while maintaining high MRI quality.
To our knowledge, no validated system currently exists to evaluate MRI diagnostic
quality. The proposed quality scoring system presented herein is intended to serve
as a starting point for future investigations. We recognize that quality score
A—no or minimal limitations—has different practical implications for
each modality, as US remains a screening modality whereas MRI is still considered a
diagnostic modality. Recently, both image quality and adequacy, referring to the
notion of whether the clinical question could be answered as determined by the
interpreting radiologist, have been described for diagnostic imaging. Quality and
adequacy of an examination will depend on the clinical task (ie, surveillance or
diagnosis), the sequences included, and how the sequences were weighted. For
example, a high-quality examination will generally be diagnostically adequate, but a
low-quality examination may still be adequate depending on the clinical task and
which sequences were impacted by reduced quality (21). Future studies should refine definitions of adequacy and validate
quality scoring systems, particularly for screening, while also considering how scan
duration and other factors affect adherence.
In the United States, the rising incidence of HCC linked to metabolic
dysfunction–associated steatotic liver disease and obesity, both of which may
impair US sensitivity, is reshaping the screening landscape (9,12). Studies show that
poor US quality scores drastically reduce sensitivity for HCC detection (US-A or
US-B, 75%–79.6% vs US-C, 21.7%–27.3%) (32,33). A study from the United
States (n = 2053) found that patients with metabolic
dysfunction–associated steatotic liver disease had the highest rate of severe
US limitations (9.5%) compared with other liver disease etiologies (9). Poor liver visualization at US may be
attributed to not only obesity, but also to male sex, nonviral etiologies of chronic
liver disease, and decompensated cirrhosis, with up to one-third of US examinations
rendered suboptimal or inadequate (9,10). The rate of severe limitations at US
(US-C) in our study was relatively low (14%) compared with other studies (10), possibly due to a predominance of
compensated cirrhosis despite a high prevalence of obese participants.
Obesity was the sole factor associated with reduced image quality across both
modalities at univariable analysis, consistent with prior studies, but was not
confirmed at multivariable analysis, possibly due to study sample size. A previous
study showed that obesity was a strong and independent predictor for severe
visualization limitation at US (OR, 5.1 [95% CI: 1.1, 23.1]; P =
.03), after adjusting for age, sex, and ethnicity (compared with the univariable OR
of 2.50 [95% CI: 1.00, 6.00; P = .04] in our study) (34). The observation that the univariable OR
for MRI was notably higher than that for US suggests that MRI may be more
susceptible to obesity-related degradation than previously appreciated. Although,
quality was not affected in the majority of cases in MRI as opposed to US. Further
studies to clarify the relative robustness of each MRI sequence in patients with
higher BMI is warranted, as it is possible obesity impacts the different sequences
differently and could be ameliorated by using a multiparametric approach. On the
contrary, a recent case-control study linked US quality limitations to obesity (OR,
2.1 [95% CI: 1.3, 3.3]) and metabolic dysfunction–associated steatotic liver
disease (OR, 2.0 [95% CI: 1.1, 3.4]), but not to dynamic contrast-enhanced MRI using
extracellular contrast material, as factors impacting HCC detection (32). A prospective study suggested US adequacy
should be reported in all clinical examinations and alternative modalities
considered when suboptimal (34).
Our data show that MRI quality remained high in obese participants, even in those
with reduced US quality, such that patients with US-B or US-C are likely to have
diagnostic MRI surveillance examinations. Our proposed MRI global quality score
incorporates several factors that can negatively impact quality, including motion
(especially during the AP and HBP), DWI artifacts, and liver uptake of gadoxetate
(reflecting liver function). The very low rate of transient severe motion (1.2%),
which is below previously reported rates (19), could be explained by the high prevalence of compensated liver disease,
low ascites prevalence, and use of multi-AP imaging. Only 0.8% of participants had
severe HBP limitations, lower than the 15%–19% reported elsewhere (34,35).
Limitations of our study include the academic setting and expertise of participating
centers, which may limit generalizability and could overestimate the feasibility and
comparative performance of MRI in broader clinical practices. Our study sample only
includes participants who were eligible for both MRI and US, which introduces a
selection bias toward individuals without contraindications to MRI. Our analysis
reflects a per-protocol assessment among participants eligible for MRI-based
surveillance rather than a full intention-to-treat population, as inclusion required
successful completion of both modalities. Data are preliminary (two of five sites),
and MRI assessments were performed by single observers. The dataset used for
observer calibration and/or training and interobserver assessment was separate from
the present study, which limits confirmation of interobserver consistency across the
study sample and imaging spectrum. Broader assessment would strengthen validation of
the proposed scoring system. US quality was extracted from the clinical report, and
we did not assess the experience or reproducibility of the sonographers, which is
associated with US examination quality (9). We
recognize potential intersite variability due to differences in sonographer
technique, equipment, and interpretation. However, in real-world surveillance
settings, variability in technique and sonographer expertise is unavoidable. Future
work would benefit from centralized, blinded US review to enhance comparability and
reduce bias. The US LI-RADS quality scores (A, B, or C) are subjective, whereas our
MRI system is semiquantitative and uses subjective scales. We attempted to address
this limitation by providing explicit operational definitions and anchoring the
criteria in clinically relevant domains, such as motion, liver parenchymal signal,
and artifacts. Our study assumes equivalence of “A,”
“B,” and “C” quality scores between MRI and US, which is
challenged by differences in modality, scoring criteria, and multiparametric MRI
sequences that may impact score distribution. To address this, we applied the same
three-tier scale (A: no or minimal limitations, B: moderate limitations, C: severe
limitations) across both imaging modalities. We recognize these fundamental
differences in methodology; however, this analysis provides a useful starting point
for further exploration. Most of our study sample undergoing HCC surveillance
consisted of participants with compensated liver disease, potentially limiting
generalizability to patients with decompensated disease (36). Although the current study did not include a formal cost
analysis, US is substantially less expensive and more widely accessible than MRI,
which may impact modality selection, particularly in resource-limited settings in
the United States and across global health care settings. MRI is more costly and
time-intensive than US, though gadoxetate-enhanced MRI can be cost-effective in
select populations, depending on HCC incidence, examination costs, and liver disease
etiology (20). Future work in this
multicenter study will explore cost-effectiveness more directly.
In conclusion, in this prospective bicenter study in the United States, most
participants undergoing hepatocellular carcinoma (HCC) screening had excellent
gadoxetate-enhanced MRI quality even when US image quality was reduced, but quality
was adversely affected by obesity with both imaging modalities (at univariate
analysis) and by Child-Pugh score with MRI only (at univariate and multivariate
analysis). There was no evidence of an association between MRI and US quality
scores, even as obesity was associated with reduced quality for both imaging
modalities at univariable analysis. Obese participants with US-C liver visualization
scores are likely to have adequate HCC surveillance with MRI. Further work is needed
to determine how image quality impacts diagnostic accuracy for HCC detection and to
define the most optimal imaging surveillance approach in the United States.

Supplemental Files
Tables S1-S2Conflicts of Interest