Clinical Application of Liver Imaging Reporting and Data System for Characterizing Liver Neoplasms: A Meta-Analysis

The Liver Imaging Reporting and Data System (LI-RADS) is a comprehensive system for standardizing liver imaging in patients at risk of developing hepatocellular carcinoma (HCC). We aimed to determine the diagnostic performance of LI-RADS category 5 (LR5) for diagnosing HCC and LI-RADS category M (LRM) for characterizing other non-HCC malignancies (OM) using contrast-enhanced ultrasound (CEUS) and computed tomography (CT)/magnetic resonance imaging (MRI). Multiple databases were searched for articles evaluating the diagnostic accuracy of CEUS LI-RADS and/or CT/MRI LI-RADS. A random-effects model was adopted to synthesize the summary estimates of the diagnostic accuracy of LR5 for diagnosing HCC and LRM for characterizing OM using CEUS and CT/MRI. The pooled sensitivity and specificity of CEUS LR5 for the diagnosis of HCC were 69% and 93%, respectively. The pooled sensitivity was 67% and the specificity, 93% of CT/MRI LR5 for HCC diagnosis. There was no significant difference between the overall diagnostic accuracy for HCC diagnosis of CEUS LR5 and that of CT/MRI LR5 in terms of diagnostic odds ratio (DOR) (p = 0.55). The sensitivity was 84% with a specificity of 90% in the CEUS LRM for characterizing OM, while the sensitivity and specificity of CT/MRI LRM for characterizing OM was 63% and 95%. The DOR of CEUS LRM for characterizing OM was higher than that of CT/MRI LRM without significant difference (50.59 vs. 36.06, p = 0.34). This meta-analysis indicated that CEUS LI-RADS is qualified to characterize HCC and OM and may provide complementary information on liver nodules to CT/MRI LI-RADS.


Introduction
Hepatocellular carcinoma (HCC) is a major cause of morbidity and mortality of cancer worldwide [1]. Unlike other malignancies, HCC can be non-invasively diagnosed in the absence of histological assessment. Imaging has been recommended as a vital diagnostic tool for HCC diagnosis in patients at high risk for developing HCC [2]. Hence, dependable imaging is critical.
CT and MRI are recommended as the first-line diagnosis methods for HCC diagnosis because of their powerful differential diagnosis of liver neoplasms [2]. In order to standardize reporting and data collection of CT and MRI for HCC diagnosis, the American College of Radiology (ACR) released the CT/MRI Liver Imaging Reporting and Data System (LI-RADS) in 2011 [3]. The three latest versions of CT/MRI LI-RADS were released in 2014, 2017, and 2018. An updated version in 2018 was finally adopted by the American Association for the Study of Liver Disease (AASLD) [4].
Along with the development of sonicated microbubbles, contrast-enhanced ultrasound (CEUS) is able to characterize focal liver lesions commendably. However, several The overall risk of bias of the 39 studies was low to moderate, shown in Supplementary Figure S1. Patient selection was the most significant source of bias. Thirteen studies were identified as being at high risk of bias in the domain of patient selection because they did not avoid a case-control design and/ or enroll lesions diagnosed by composite clinical reference standards. Studies that employed LR5 of another imaging modality without follow-up as a reference standard were considered at high risk of bias in the domain of reference standard.

Quality Assessment
The overall risk of bias of the 39 studies was low to moderate, shown in Supplementary Figure 1. Patient selection was the most significant source of bias. Thirteen studies were identified as being at high risk of bias in the domain of patient selection because they       The pooled sensitivity was 63% (95% CI, 57-69%; I 2 = 59.33%; p < 0.01) and the specificity, 95% (95% CI, 93-97%; I 2 = 85.68%; p < 0.01) of CT/MRI LRM for characterizing OM ( Figure 3B).  Figure 3C).

Meta-Regression Analysis
Meta-regression analyses of the sensitivity and specificity of CEUS LR5 for diagnosing HCC and CEUS LRM for characterizing OM are demonstrated in Supplementary Table S4. Among six covariates, country, average lesion size and proportion of OM were the factors that contributed to the heterogeneity of the sensitivity of CEUS LR5 for diagnosing HCC. Regarding the reference standard, the specificity of CEUS LR5 for diagnosing HCC was higher for studies using a mixed reference standard as a reference standard than for those that employed only pathology (96% vs. 89%; p = 0.06). Country, study design, proportion of OM, and reference standard were the factors that significantly influenced the heterogeneity of the specificity of the CEUS LRM category for characterizing OM.
Meta-regression analyses of the sensitivity and specificity of the CT/MRI LR5 category for diagnosing HCC and the CT/MRI LRM category for characterizing OM were also performed (Supplementary Table S5). We noted that the pooled sensitivity was 67% (95% CI, 62-72%) and the specificity, 93% (95% CI, 91-96%) of LR5 for diagnosing HCC with CT/MRI LI-RAD v2017/2018, superior to those with v2014 (sensitivity, 65%, 95% CI, 56-74%; specificity, 92%, 95% CI, 88-96%; p = 0.05). As for average size, the sensitivity of CT/MRI LI-RADS for diagnosing HCC and characterizing OM was reduced when small lesions were enrolled (Sensitivity of LR5 for diagnosing HCC, 61% vs. 68%; Sensitivity of LRM for characterizing OM, 57% vs. 64%), similar to the findings observed in CEUS studies. Besides, the index test was one of the factors that significantly influenced the heterogeneity of the sensitivity and specificity for both HCC and non-HCC malignancies with CT/MRI LI-RADS.

Publication Bias
No obvious publication bias was identified through the funnel plot and Egger's test (p > 0.05), shown in Supplementary Figure S2.

Discussion
The meta-analysis included more than 5000 lesions in each LI-RADS. A pooled analysis revealed a sensitivity of 69% and a specificity of 93% of CEUS LR5 and a sensitivity of 67% and a specificity of 93% of CT/MRI LR5 for diagnosing HCC. This study also indicated that the pooled sensitivity and specificity for characterizing OM with respect to CEUS LRM were 84% and 90%, respectively, and those of CT/MRI LRM were 63% and 95%.
Our meta-analysis found that the pooled sensitivity and specificity for diagnosing HCC of CEUS LR5 was 69% and 93%, respectively, which was quite similar to those summary estimates of CT/MRI LR5 (sensitivity 67% and specificity 93%). A study conducted by Shin et al. [16] suggested that the diagnostic performances for diagnosing HCC of LI-RADS with CEUS and CT/MRI can be compared. In our study, there was no statistical difference between CEUS LR5 and CT/MRI LR5 in the overall diagnostic accuracy for diagnosing HCC (p = 0.55), but a specificity of 93% for HCC diagnosis of LR5 was lower than that preconceived in the LI-RADS algorithm (definitely HCC).
Patients with ICC and HCC-CCA have a poorer prognosis and fewer treatment options than those patients with HCC. Therefore, distinguishing OM from HCC will be of great significance for making clinical decisions. In our meta-analysis, CEUS LRM showed high sensitivity (84%) and high specificity (90%) for characterizing OM, while moderate sensitivity (63%) and high specificity (95%) were observed in CT/MRI LRM. The overall diagnostic accuracy of CEUS LRM was higher than that of CT/MRI LRM without significant difference (p = 0.34). In other words, compared to CT/MRI, CEUS did not increase the risk of misdiagnosing OM as HCC. This finding is not in agreement with the results of some previous studies which indicated CEUS was associated with a high risk of misdiagnosis [7].
CEUS is different from CT/MRI in the following aspects [59]: (a) physicochemical properties of contrast agent used, (b) method to obtain images (for CEUS, dynamic realtime imaging; for CT/MRI, static imaging) and (c) criteria adopted to assign the LI-RADS category to a liver nodule. Based on the above, CEUS may provide additional information of focal liver nodules to CT/MRI. We suggested that CEUS is useful in assisting to provide an accurate diagnosis for patients at risk of HCC and a multimodality approach might be needed to improve the diagnostic performance of characterizing HCC and OM.
Regarding the baseline characteristics of the included studies, we noted that most were from Asian countries, followed by European and North American countries. Although the epidemiological features and imaging characteristics of HCC in non-Asian countries were different from that in Asian countries, CEUS LI-RADS and CT/MRI LI-RADS introduced by the American association were also suitable to be applied in Asian countries where HBV is endemic. Patients with HBV infection are at risk for HCC even without cirrhosis. In a cirrhotic liver, we need to differentiate HCC from regenerating nodules, and for patients with HBV infection in the absence of cirrhosis, we do not need to do so. This might lead to superior diagnostic performance for HCC. However, the prevalence of cirrhosis in the whole sample was lower than that in western countries, so the generalizability of these results should be further assessed. There are only a limited number of studies evaluating the diagnostic performance of CEUS LI-RADS.The main reasons are as follows: (a) The CEUS LI-RADS was officially issued in 2017, while CT/MRI LI-RADS was first released in 2011 and multiple revisions were updated after that. In other words, CEUS LI-RADS has a short time of application worldwide in comparison to CT/MRI LI-RADS. (b) Some centers adopted other CEUS-based standardized algorithms (such as ESCULAP, EFSUMB) in clinical practice instead of LI-RADS. Among 39 included records, only eight were prospective studies, eight were multicenter studies and twenty-nine employed a cohort study design. More prospective, multicenter cohort researches that focus on the LI-RADS algorithm are expected to be published in the future.
QUADAS-2 was used to assess the methodological quality of the included studies. More than half of the studies had at least one domain at high risk of bias. In our study, case-control studies were viewed to be inferior to those cohort studies since the percentage of HCC and OM depended on the researchers. Using pathology as a reference standard exclusively might introduce verification bias because lesions receiving pathological assessment are more inclined to be malignant and be assigned to higher LI-RADS categories. There is likely to be incorporation bias for studies employing LR5 of another imaging modality without follow-up as a reference standard [14].
To a certain extent, there were clinical and methodological similarities between studies because of cautious study selection and methodological quality assessment, but there was still statistical heterogeneity between the included studies. The meta-regression analysis showed that the specificity of the CEUS LR5 for diagnosing HCC in the studies employing merely pathology as a reference standard was significantly lower than those employing a mixed reference standard (89% vs. 96%, p = 0.03). A similar finding was observed in the CT/MRI studies. HCC with typical imaging features can be easily diagnosed by composite imaging, whereas atypical HCC was difficult to be definitely diagnosed by imaging and should be biopsied. Studies using only a pathological reference standard might contain more atypical HCC than those using a mixed reference standard, which may explain their lower specificity for characterizing HCC. We also noted that when small lesions were enrolled, the sensitivity for diagnosing HCC and characterizing OM was reduced both at CEUS and CT/MRI. Small liver nodules usually possess atypical imaging features, so that it remains a challenge to distinguish small HCC and OM from other focal liver lesions, especially under the background of cirrhosis.
Comparing CT/MRI LI-RADS v2014, v2017 and v2018 made major revisions to applicable patients, diagnostic categories, threshold growth definition, LRM criteria, and ancillary features. The pooled sensitivity was 67% and the specificity was 93% of LR5 for diagnosing HCC with CT/MRI LI-RAD. v2017/2018 was superior to those with v2014 (sensitivity, 65%; specificity, 92%; p = 0.05). It seemed that the revisions significantly improved the diagnostic performance for diagnosing HCC. The index test significantly influenced the heterogeneity of the sensitivity and specificity for both HCC and non-HCC malignancies with CT/MRI LI-RADS. So we performed three meta-analyses that compared the different techniques (CEUS, CT, MRI) two by two after removing four studies [30,33,40,41] which didn't evaluate the diagnostic accuracy at CT and MRI imaging separately (Supplementary Table S6).
We tried to answer the question of whether CEUS is qualified to characterize HCC and OM accurately, which were closely relevant to the management of liver nodules. There were some limitations. First, apparent heterogeneity was identified in our meta-analysis. To seek the covariates that contributed to the heterogeneity, a meta-regression analysis was performed, but the heterogeneity still cannot be ignored. Second, an adjusted indirect comparison was used instead of direct comparison, mainly due to insufficient evidence comparing the diagnostic accuracy of CEUS LI-RADS and CT/MRI LI-RADS head-tohead. A study reported that adjusted indirect comparison might be less biased than direct comparison [60]. Finally, we did not perform subgroup analyses based on major and ancillary features of the LI-RADS because of insufficient data. Thus, our results should be confirmed by more evidence. Considering the limitations, more high-quality direct comparisons between CEUS LI-RADS and CT/MRI LI-RADS are required to update this meta-analysis. At present, we have been working on a multicenter study comparing the diagnostic accuracy of CEUS LI-RADS and CT/MRI LI-RADS for characterizing HCC and OM directly with a large sample size. We also keep a concurring opinion with the study [14] that an individual participant meta-analysis is needed.
In conclusion, both CEUS LR5 and CT/MRI LR5 showed moderate sensitivity and high specificity for diagnosing HCC. CEUS LRM showed high sensitivity and high specificity for characterizing OM, while CT/MRI LRM showed moderate sensitivity and high specificity.
Supplementary Materials: The following are available online at https://www.mdpi.com/2075-4418/11/2/323/s1, Figure S1. Methodological quality assessment using the Quality Assessment of Diagnostic Accuracy Studies 2 tool; Figure S2. Deek's funnel plot and Egger's test to evaluate the publication bias of (A) studies evaluating the diagnostic performance of CEUS LR5 for HCC diagnosis, (B) studies evaluating the diagnostic performance of CT/MRI LR5 for HCC diagnosis, (C) studies evaluating the diagnostic performance of CEUS LRM for characterizing OM, and (D) studies evaluating the diagnostic performance of CT/MRI LRM for characterizing OM. CEUS, contrastenhanced ultrasound; HCC, hepatocellular carcinoma; OM, other non-HCC malignancies; LR5, Liver Imaging Reporting and Data System category 5; LRM, Liver Imaging Reporting and Data System category M; ESS, effective sample sizes; p > 0.05, no obvious publication bias was identified; Table S1. Details of search strategy; Table S2. Data form; Table S3. Details of excluded studies; Table S4. The results of meta-regression of the sensitivity and specificity of the CEUS LR5 for diagnosing HCC and the CEUS LRM for characterizing OM; Table S5. The results of meta-regression of the sensitivity and specificity of the CT/MRI LR5 for diagnosing HCC and the CT/MRI LRM for characterizing OM; Table S6. The diagnostic performance of LR5 for diagnosing HCC and LRM for characterizing OM with CEUS, CT and MRI separately.