Quality Assessment of Ultrasound and Magnetic Resonance Imaging for Hepatocellular Carcinoma Surveillance: A Systematic Review and Meta-Analysis

Introduction: To achieve early detection and curative treatment options, surveillance imaging for hepatocellular carcinoma (HCC) must remain of quality and without substantial limitations in liver visualization. However, the prevalence of limited liver visualization during HCC surveillance imaging has not been systematically assessed. Utilizing a systematic review and meta-analytic approach, we aimed to determine the prevalence of limited liver visualization during HCC surveillance imaging. Methods: MEDLINE and Embase electronic databases were searched to identify published data on liver visualization limitations of HCC surveillance imaging. An analysis of proportions was pooled using a generalized linear mixed model with Clopper-Pearson intervals. Risk factors were analysed using a generalized mixed model with a logit link and inverse variance weightage. Results: Of 683 records, 10 studies (7,131 patients) met inclusion criteria. Seven studies provided data on liver visualization limitations on ultrasound (US) surveillance exams: prevalence of limited liver visualization was 48.9% (95% CI: 23.5–74.9%) in the overall analysis and 59.2% (95% CI: 24.2–86.9%) in a sensitivity analysis for cirrhotic patients. Meta-regression determined that non-alcoholic fatty liver disease was associated with limited liver visualization on US. Four studies provided data for liver visualization limitations in abbreviated magnetic resonance imaging (aMRI), with inadequate visualization ranging from 5.8% to 19.0%. One study provided data for complete MRI and none for computed tomography. Conclusion: A substantial proportion of US exams performed for HCC surveillance provide limited liver visualization, especially in cirrhosis, which may hinder detection of small observations. Alternative surveillance strategies including aMRI may be appropriate for patients with limited US visualization.


Abstract
Introduction: To achieve early detection and curative treatment options, surveillance imaging for hepatocellular carcinoma (HCC) must remain of quality and without substantial limitations in liver visualization. However, the prevalence of limited liver visualization during HCC surveillance imaging has not been systematically assessed. Utilizing a systematic review and meta-analytic approach, we aimed to determine the prevalence of limited liver visualization during HCC surveillance imaging. Methods: MEDLINE and Embase electronic databases were searched to identify published data on liver visualization limitations of HCC surveillance imaging. An analysis of proportions was pooled using a generalized linear mixed model with Clopper-Pearson intervals. Risk factors were analysed using a generalized mixed model with a logit link and inverse variance weightage. Results: Of 683 records, 10 studies (7,131 patients) met inclusion criteria. Seven studies provided data on liver visualization limitations on ultrasound (US) surveillance exams: prevalence of limited liver visualization was 48.9% (95% CI: 23.5-74.9%) in the overall analysis and 59.2% (95% CI: 24.2-86.9%) in a sensitivity analysis for cirrhotic patients. Meta-regression determined that non-alcoholic fatty liver disease was associated with limited liver visualization on US. Four studies provided data for liver visualization limitations in abbreviated magnetic resonance imaging (aMRI), with inadequate visualization ranging from 5.8% to 19.0%. One study provided data for complete MRI and none for computed tomography. Conclusion: A substantial proportion of US exams performed for HCC surveillance provide limited liver visualization, especially in cirrhosis, which may hinder Introduction Hepatocellular carcinoma (HCC) is the sixth most common cancer but the third leading cause of cancer death worldwide [1]. The disease stage at diagnosis is a major determinant of outcomes [2]. Optimal surveillance imaging for HCC relies on good-quality imaging without major limitations in liver visualization [3]. This is critical for early detection and curative treatment of HCC [4]. Detection of HCC at an early stage may facilitate curative treatment options such as surgical resection or liver transplantation, which are associated with 5-year survival rates exceeding 70% [5,6]. Conversely, those with advanced HCC at diagnosis have poor outcomes, with a median survival of 10-20 months among those receiving systemic therapy [2].
Major society guidelines recommend semi-annual ultrasound (US) with or without alpha-fetoprotein (AFP) for HCC surveillance [7,8]. However, a meta-analysis determined that the sensitivity of US for detecting earlystage (defined as within Milan criteria) HCC was less than 50% [9]. US is operator-dependent, and limited liver visualization during US surveillance is associated with poorer surveillance test performance, missed HCC, and unnecessary diagnostic investigations [3,10]. Computed tomography (CT) and complete magnetic resonance imaging (MRI) have been utilized for HCC screening but are limited by radiation and long exam times, respectively [11]. The limitations of US, CT, and conventional MRI have led several investigators to explore hepatobiliary phase-abbreviated gadoxetate-enhanced magnetic resonance imaging (referred to hereafter as abbreviated magnetic resonance imaging [aMRI]). aMRI utilizes a variety of sequences, including diffusion-weighted imaging, T2-weighted and T1weighted sequences, typically utilizing only a third of the sequences typically performed in a full diagnostic liver MRI, resulting in exam times that are potentially less than 10 min [12,13]. Existing systematic reviews and metaanalyses have focused mainly on the test performance of imaging modalities to detect HCC [14,15]. Less research has been performed on surveillance imaging quality, even though it is known to be a major determinant of performance, and the prevalence of limited liver visualization during HCC surveillance imaging has not been systematically evaluated. Herein, we performed a systematic review and meta-analysis to estimate the prevalence of HCC surveillance imaging with limited liver visualization and to compare the prevalence associated with different modalities and approaches.

Search Strategy and Inclusion Criteria
A search was conducted on MEDLINE and Embase electronic databases from inception to March 3, 2022. The current systematic review and meta-analysis were synthesized per the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRIS-MA) statement [16,17]. Key search terms included "hepatocellular carcinoma," "ultrasound," "visualization," "Computed tomography," and "Abbreviated MRI." The full search strategy is included in online supplementary Material 1 (for all online suppl. material, see https://doi.org/10.1159/000531016). The references were then compiled with Endnote X20, and duplicates were removed. Additionally, references from previous meta-analyses on HCC surveillance were also screened to ensure the comprehensiveness of the search [4,13].

Eligibility and Selection Criteria
Two authors (J.Q. and D.J.H.T.) independently screened abstracts and conducted full-text reviews to check the eligibility for inclusion, and disputes were resolved by obtaining the consensus of a senior author (D.Q.H.). Only original articles including retrospective or prospective cohort studies, randomized control trials, or case-control studies were considered for inclusion, with reviews, commentaries, letters to editors, and editorials being excluded. If multiple studies utilized data from the same database, we only included data from the most updated study.

Definitions
For US, we adopted the standardized LI-RADS definitions of limited liver visualization, which are scored as B (moderate limitations) or C (severe limitations). Conceptually, these are limitations that may obscure small observations (LI-RADS US visualization score B) or that may lead to a significant reduction in focal liver lesion detection (score C) [18,19]. There are not yet standardized definitions for limited visualization on other modalities. Hence, definitions used by the included studies for liver visualization in aMRI and MRI were heterogenous (see Results). This precluded a pooled analysis, and we reported our findings for MRI in the form of a systematic review. As there were no included CT studies, there were no CT-specific definitions.

Endpoints
The primary endpoint of this analysis was the prevalence of HCC surveillance imaging with limited liver visualization. Secondary endpoints were the risk factors associated with inadequate visualization quality.

Quality Assessment
The Joanna Briggs Institute (JBI) Critical Appraisal Tool was used for the quality assessment of included articles [20] assessment rates the quality of cohort studies based on the appropriateness of sample frame, sampling method, adequacy of sample size, data analysis, methods for identification and measurement of relevant conditions, statistical analysis, and response rate adequacy. Online supplementary Material 2 summarizes the quality assessment scores for included articles.

Statistical Analysis
All analyses were conducted in RStudio (version 5.1.3). Statistical heterogeneity was assessed via I 2 and Cochran's Q test values, where I 2 values of 25%, 50%, and 75% represented low, moderate, and high degrees of heterogeneity, respectively [21,22]. A Cochrane Q test of p value ≤0.10 was considered significant for heterogeneity [21]. A random-effect model was used in all analyses regardless of heterogeneity measures as evidence has demonstrated more robust effect estimates with random-effect compared to fixed-effect models [23,24]. An analysis of proportions was pooled using a generalized linear mixed model with Clopper-Pearson intervals [25,26]. To assess for risk factors, a generalized mixed model with a logit link and inverse variance weightage was used to derive the odds ratio for risk factors where sufficient data were available for pooled analysis [27]. A sensitivity analysis based on patients with cirrhosis was conducted for the risk factors associated with limited liver visualization. We conducted a systematic reporting of the available data for outcome measures from aMRI as it was deemed too heterogenous for a pooled analysis. A p value <0.05 was considered the threshold for statistical significance.

Summary of Included Articles
The initial search from MEDLINE and Embase yielded a total of 683 articles. After the removal of 262 duplicates during the sieving of titles and abstracts, 421 papers remained for abstract screening, and a total of 10 papers were included in the meta-analysis (Fig. 1). The studies included were conducted in Canada [28], South Korea [19,29], and the USA [3,[30][31][32][33][34][35]. A total of 7,131 patients were included in our analysis, of which 3,649 patients had cirrhosis. A total of seven studies provided data on liver visualization limitations in US [3,28,29,[31][32][33][34], four in aMRI [19,30,31,35], and one in complete MRI [19]. One study provided data on the quality of liver visualization for both US and aMRI [31], and one study provided data for both aMRI and complete MRI [19]. The quality of all included studies, based on the Joanna Briggs Institute Critical Appraisal Checklist, was assessed to have a low risk of bias. A summary of the included articles is included in online supplementary Material 2.

Meta-Regression of Risk Factors for Limited Liver Visualization on US Conducted for HCC Surveillance
Non-alcoholic fatty liver disease (NAFLD) was associated with higher odds of limited visualization quality on US (OR: 1.24; 95% CI: 1.20-1.29; p < 0.01). Study-level factors including age, sex, BMI, and other aetiologies of liver disease (hepatitis C, hepatitis B, alcohol-associated liver disease) were not significantly associated with increased odds of limited liver visualization ( Table 2).

Prevalence of Limited Visualization in MRI Performed for HCC Surveillance
A total of 4 studies provided data for the proportion of limited liver visualization in gadoxetate-enhanced aMRI scans performed for HCC surveillance, one for complete MRI, and none for CT ( Table 3). The definitions utilized for limited visualization in aMRI were heterogenous (Table 3), precluding a pooled analysis. Brunsing et al. [35] conducted a retrospective study of 330 consecutive patients (93.3% with cirrhosis, 6.7% chronic hepatitis B without cirrhosis) that underwent gadoxetate-enhanced aMRI for HCC screening. Among patients with cirrhosis, the most common aetiologies of cirrhosis were hepatitis C virus (41.2%) and alcohol (19.5%). The authors of this study determined that 5.8% of aMRI were of "inadequate quality." Park et al. [19] conducted a retrospective analysis of a prospective cohort of 382 patients with cirrhosis (72.3% hepatitis B, 12.0% alcohol) that underwent complete gadoxetic acid-enhanced MRI and US. aMRI was simulated using the collected images from the complete gadoxetic acid-enhanced MRI scans. They determined that liver visualization was limited (i.e., small nodules might be obscured or missed) in 33.7% of US, 10.0% of aMRI, and 11.1% of complete gadoxetic acid-enhanced MRI scans. Kobi et al. [30] conducted a retrospective study of patients with chronic liver disease who underwent HCC screening with complete Gd-EOB-DTPAenhanced MRI. The authors determined that 14.8% of the hepatobiliary phase phases were "suboptimal." Prevalence of Limited Visualization by Surveillance Imaging in Studies That Compared Two or More Modalities Head-to-Head Two studies [19,31] compared limited US and aMRI head-to-head in the same cohort. In both studies, the prevalence of limitations was numerically higher for US than for aMRI (34% vs. 10% and 35% vs. 19%).

Main Findings
In this systematic review and meta-analysis, we determined that 45% of US scans performed in patients with cirrhosis provided limited visualization of the liver. In a  Vis B, visualization score B; Vis C, visualization score C; 95% CI, 95% confidence interval; I 2 , level of heterogeneity. sensitivity analysis for cirrhosis, 52% of US scans provided limited liver visualization. Meta-regression determined that NAFLD was associated with an increased risk of limited visualization on US. By contrast, between 6% and 19% of aMRI scans provided limited visualization. Only one study reported data for limited visualization in complete gadoxetate-enhanced MRI performed for HCC surveillance and reported a similar rate of limited visualization in complete MRI (11%) and aMRI (10%).
These data have important clinical implications. We determined that a substantial proportion of US performed for HCC surveillance provides limited liver visualization.
Limited visualization on US for HCC surveillance is associated with an increased risk of false-negative results, with a reduction in sensitivity of more than 50% in the setting of severely limited visualization [3,29]. Patient factors including increased BMI and male sex and procedure-related factors including rib shadowing and breath motion abnormalities have previously been associated with inadequate US image quality [36]. Moreover, the presence of cirrhosis presents additional challenges to imaging quality due to disruption of the liver parenchyma and increased false positives from distorted benign lesions being misinterpreted as HCC [37]. Since the risk of missed HCC is high when there is limited liver visualization, we propose that the US LI-RADs score should be routinely documented in patients that undergo US for HCC surveillance, and care providers should either repeat the US or consider alternative surveillance strategies, such as aMRI, for patients with limited US visualization [38,39]. In addition, MRI-based screening modalities may offer increased sensitivity for smaller HCC lesions, although sensitivity is lower in lesions <2 cm [13,40,41]; however, access to MRI is an issue, and further cost-effectiveness studies are needed to determine optimal HCC screening protocols. The use of contrast in MRI may have the added advantage of providing further granularity for treatment options and prognosis for HCC in the indeterminate stage and in those with hepatitis C virusrelated HCC [42,43], but further data are required to validate this hypothesis. Furthermore, the large proportion of US scans with limited visualization for HCC surveillance highlights the need for biomarkers to detect early HCC [44,45].
Secondly, we determined that NAFLD was associated with limited visualization on US. With the rising global burden of NAFLD-related HCC, there is an unmet need to develop better surveillance strategies for patients with NAFLD [46,47]. In particular, health care providers, clinical practice guidelines, and patients should understand that US may provide limited liver visualization in the context of NAFLD. Emerging data suggest that aMRI may be useful for HCC surveillance in NAFLD cirrhosis, but larger studies are required [31]. It is notable that only one study prospectively assessed visualization quality for HCC surveillance, and this study determined a substantially higher proportion of limited visualization quality compared to retrospective studies [31]. This suggests that dedicated imaging quality assessment may detect a higher rate of limited visualization, and additional prospective studies are required.
There are currently limited data on how visualization related to US test performance. A recent study by Chong et al. [3] determined that limited visualization quality on US was associated with substantially higher odds of false-positive and false-negative results. The current study highlights the large proportion of US with limited visualization quality, which may contribute to the limited sensitivity (<50%) of US for detecting early HCC, as determined by Tzartzeva et al. [9]. Taken together, these data highlight the need for greater awareness among care providers of visualization quality on US performed for HCC surveillance and alternative strategies.

Strengths and Limitations
The current study is the first meta-analysis to evaluate objective scoring of visualization quality in HCC surveillance imaging. However, it is not without limitations. The impact of limited exams was not assessed with regard to missed HCC. The impact of sonographer experience was not available to determine whether greater expertise/experience may improve outcomes [48]. Variations in body habitus also contribute to heterogeneity and could not be quantified due to a paucity of data in the included articles. Additionally, the semi-quantitative scoring of liver visualization using LI-RADS may suffer from interreader variability, while standardized may not be entirely reliable between studies [49,50]. There were limited studies that provided data for visualization quality for aMRI, only one for complete MRI, and no studies for CT. The definitions used for defining limited visualization in aMRI were heterogenous, and a consensus, standardized definition is required. We were unable to investigate the effect of tumour size on visualization quality due to a lack of data, although previous studies have demonstrated reduced sensitivity of imaging techniques for the detection of small lesions from 10 to 20 mm, especially for lesions less than 10 mm [51][52][53]. Finally, there were insufficient studies to perform subgroup analysis among patients without cirrhosis and by aetiology. However, we performed meta-regression and determined that NAFLD was associated with a higher risk of limited visualization.

Conclusion
In summary, we determined that around half of US imaging performed for HCC surveillance provides limited visualization of the liver, which is known to reduce sensitivity for HCC. Visualization is worse in cirrhosis and in patients with NAFLD. Visualization quality should be documented for all patients who undergo US for HCC surveillance. Alternative surveillance strategies, such as aMRI, should be developed for patients with limited US visualization quality.

Funding Sources
No funding was required for this study.