Introduction

Hepatocellular carcinoma (HCC) is the sixth most prevalent cancer and the third most frequent cause of cancer-related death worldwide [1,2,3]. Patient prognosis can achieve a 5-year survival rate of higher than 50% if the HCC is diagnosed at an early stage; especially those within Milano criteria (up to three nodules ≤ 3 cm) may be curable and have a desirable prognosis [2]. More accurate detection of HCC at the early stage (≤ 3 nodules ≤ 3 cm each patient) may reduce the risk of tumor recurrence [4, 5]. Therefore, it is important to effectively improve the sensitivity of diagnosis for HCC at the early stage. HCC can be diagnosed by dynamic contrast-enhanced MRI based on typical imaging feature, hyperintensity on arterial phase, and hypointenstity on portal venous or delayed phases [6, 7]. However, there are still some challenges in the detection and characterization of small HCC because lacking typical imaging feature may lead to lower sensitive (44–62%) for small HCC [8], failing to identify approximately 40% small HCC cases [9, 10].

The introduction of liver-specific contrast agent gadoxetic acid (Gd-EOB-DTPA) provides both hemodynamic information during early dynamic phases and additional information during the hepatobiliary phase (HBP) [11,12,13,14]. Most HCC displays hypointensity on HBP, and this feature weights differently depending on the guidelines. Hypointensity on HBP is a major feature in Asian guidelines, whereas western guideline considers it to be an ancillary feature, not sufficient to make a conclusive diagnosis of HCC because of concern over the loss of specificity [7, 15, 16]. Many studies have shown that the combined interpretation of dynamic phase and hypointensity on HBP images improves the diagnostic accuracy of Gd-EOB-DTPA-enhanced MRI for detection of HCC [13, 17,18,19,20,21,22]. As a consequence, hypointensity on the HBP may be a major hallmark to detect small HCC for Gd-EOB-DTPA-enhanced MRI, particularly for atypical nodules with arterial enhancement but no portal venous or delayed phase washout.

The purpose of our study was to perform a meta-analysis to synthesize the diagnostic accuracy of combining HBP with dynamic phase in gadoxetic acid–enhanced MRI for detection of HCC measuring up to 3 cm, and compare with contrast-enhanced MRI without HBP images based on hyperintense on arterial phase followed by venous or delayed phases washout in patients with chronic liver disease. We also explored factors that may influence the diagnostic accuracies for HCC of enhanced MRI.

Materials and methods

This meta-analysis was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for diagnostic test accuracy study (PRISMA-DTA) [23].

Search strategy

Literature search was performed in several databases, including PubMed, Web of Science, Embase, and Scopus, to retrieve studies about gadoxetic acid–enhanced MRI published from inception to April 10, 2020. The comprehensive search strategy was as the following terms: hepatocellular carcinoma/HCC, MRI/magnetic resonance imaging, and gadoxetic/gadoxetate/EOB/eovist/primovist. We excluded review article, abstracts, case reports, letters, and comments. All potentially appropriate studies were retrieved for full-text assessment.

Study selection

Two radiologists (with 5 and 7 years of experience in radiology, respectively) evaluated the titles and abstracts, respectively. Disagreements between the two reviewers were resolved by face-to-face discussion. The eligible full-text articles were retrieved and evaluated by the same two reviewers.

Studies were included when they met all of the following criteria: (1) study population older than 18 years with chronic liver disease and hepatic nodules; (2) suspected HCC; (3) diagnostic accuracy of Gd-EOB-DTPA-enhanced MRI; (4) histopathologic examinations (including biopsy, resection, explant), transcatheter arterial chemoembolization, or clinical follow-up period of at least 6 months as the reference standard; and (5) articles written in English. Studies were excluded when they met at least one of the following criteria: (1) all nodules with a diameter larger than 3 cm; (2) not original research, such as reviews, meta-analysis, letters or meeting abstracts; (3) articles without HBP; (4) data not available to extract or calculate true-positive (TP), false-positive (FP), false-negative (FN), or true-negative (TN); or (5) sample size fewer than 10 lesions. Disagreements of included studies between the two reviewers were resolved by discussion with a third radiologist with 20 years of experience in radiology.

Data extraction

Two previously mentioned reviewers independently screened and extracted the relevant data (TP, FP, TN, FN) from the included studies for meta-analysis.

We recorded the following basic information of each study: primary author, year of publication, type of journal, country, type of study design (retrospective or prospective study), number of patients, number of males and females, range of age and mean age, cause of cirrhosis, lesion number, size range of lesions, and reference standard, the magnetic field strength, rate of Gd-EOB-DTPA injection, acquiring time of arterial phase, portal venous phase, delayed phase and hepatobiliary phase, section thickness of contrast-enhanced imaging, and intensity in HBP of each lesion (Table 1).

Table 1 Characteristics of Gd-EOB-DTPA-enhanced MRI of included studies
Full size table

The diagnostic algorithm of Gd-EOB-DTPA-enhanced MRI with HBP for HCC was defined as a nodule showing hyperintensity in arterial phase and hypointensity in the portal venous, delayed, or hepatobiliary phases. Without HBP, the diagnostic criteria of enhanced MRI were a nodule showing hyperintensity in arterial phase and hypointensity in the portal venous or delayed phases. Data (TP, FP, FN, TN values) for the detection of up to 3 cm nodules of enhanced MRI were extracted or calculated to reconstruct a 2 × 2 contingency table. If a study reported diagnostic accuracy data of multiple observers, we chose data of the most experienced observer.

Methodological quality and risk of bias assessment

The Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) [24] was used to evaluate the risk of bias on the study level (Supporting Figure S1).

Statistical analysis

Data extracted from included studies were divided into set 1 and set 2. The data of set 1 consisted of studies containing extractable data for Gd-EOB-DTPA-enhanced MRI with HBP. The data of set 2 were composed of studies containing extractable data for enhanced MRI without HBP.

Data for Gd-EOB-DTPA-enhanced MRI with HBP

For several studies involving exact data of TP, FP, FN, and TN from set 1, the following parameters were calculated by using a bivariate random-effects model: sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratios (DOR). All corresponding 95% confidence intervals (CIs) were calculated. A summary receiver operating characteristic (sROC) curve and the area under the sROC curve (AUC) with relevant 95% CIs were also generated. All the studies from set 1 were only combined to calculate sensitivity due to the absence of true-negative data in several studies from this set.

Data for enhanced MRI without HBP

Due to the absence of exact true-negative data for enhanced MRI without HBP in studies from set 2, only the sensitivities of set 2 studies were combined and calculated.

Subgroup analysis

Subgroup analysis was based on factors that potentially affected diagnostic sensitivity or caused heterogeneity. More details of potential factors about subgroup analysis are provided in Table 2.

Table 2 Sensitivity estimates for subgroups analyses of gadoxetic acid–enhanced MRI with and without HBP in detection of small HCC
Full size table

The z test was used to compare the summarized sensitivity between enhanced MRI with HBP and without HBP. Heterogeneity was quantified by Q and I2 [25]. The MIDAS model [26] was used to perform the bivariate random-effects model and generate the sROC curve. p < 0.05 indicate a statistically significant difference. Funnel plots were used to assess the publication bias. More detailed description of the statistical methods can be found in the supplementary materials [27, 28].

Results

Study selection

Finally, 29 studies and 3554 lesions (HCC = 2696) were included in the meta-analysis (Fig. 1). Seventeen studies reported comprehensive parameters for the diagnostic accuracy, while the other 12 studies reported only sensitivity. Among the 29 studies (group 1, 17 studies containing extractable data both for MRI with HBP and without HBP; group 2, 7 studies only containing extractable data for MRI with HBP; and group 3, 5 studies only containing extractable data for MRI without HBP), studies of group 1 plus group 2 constituted set 1 and studies of group 1 plus group 3 constituted set 2 (Table 1).

Fig. 1
figure 1

Studies excluded and included in the study

Full size image

Characteristics of studies

Characteristics of the 29 included studies are summarized in Table 1. All included studies used a single-center design. In the 17 studies [17, 21, 22, 29,30,31,32,33,34,35,36,37,38,39,40,41,42], data could be extracted to calculate the TP, FP, FN, and TN values for Gd-EOB-DTPA-enhanced MRI with HBP. In the remaining 12 studies [18, 43,44,45,46,47,48,49,50,51,52,53], only the data of sensitivity could be extracted. The size of the lesions was ranging from 2 to 30 mm for Gd-EOB-DTPA-enhanced MRI. Seven studies used a prospective design, and the other 22 used a retrospective design. Magnetic field strength was 3.0 T in thirteen studies, lower than 3.0 T in 12 studies, a combination of 3.0 T and 1.5 T in 3 studies [35, 37, 44], and one study was unclear [21]. Contrast agent injection was administered at a rate of 2 ml/s in 9 studies, slower than 2 ml/s in 19 studies, and remaining one study was unclear [21]. After contrast agent administration, enhanced MRI of early and late arterial phases were performed with a threshold of 20 s. Nine studies performed enhanced MRI in the early arterial phase, whereas 18 studies performed late arterial phase. Two remaining study lacked a concrete value about acquisition time of the arterial phase [21, 48]. The histopathologic examination (biopsy, resection, explant) was the sole reference standard for all lesions in 13 studies, and a combination of histopathology, clinical follow-up, transcatheter arterial chemoembolization, or radiofrequency ablation as the comprehensive reference standard were performed in the remaining 16 studies. Six studies were conducted outside of Asia, and 23 studies were conducted in Asia. There are 10 studies designed as case series, and another 19 studies are designed as case-control.

Quality assessment

The QUADAS-2 questions evaluated for each study are presented in Supporting Figure S1. The quality assessment results for each study are shown in Supporting Table S1 and Fig. 2. Most of the included studies (14 out of 29) were found to be at a high risk of bias for patient selection due to their retrospective design. The risk of bias for the index test was considered to generate from the awareness of clinical information about patients in reviewing MRI. A combination of histopathology and clinical follow-up as the comprehensive reference standard may result in a substantial risk of bias for the reference standard. The combinational reference standard was a considerable factor that may increase the risk of bias for flow and timing.

Fig. 2
figure 2

Quality assessment results for each study

Full size image

Overall diagnostic accuracy

Diagnostic accuracy analysis

For 17 studies, data of TP, FP, FN, and TN values could be extracted in set 1 for Gd-EOB-DTPA-enhanced MRI with HBP. The summary estimates for the detection of HCC by using Gd-EOB-DTPA-enhanced MRI with HBP showed a sensitivity of 87% (95% CI: 80%, 92%), specificity of 92% (95% CI: 85%, 96%), PLR of 11.5 (95% CI: 5.8, 22.8), NLR of 0.14 (95% CI: 0.09, 0.22), and DOR of 82 (95% CI: 36, 187). Substantial heterogeneity was exhibited by the Q test for sensitivity (p < 0.001) and for specificity (p < 0.001) (Fig. 3). The summary ROC curve of Gd-EOB-DTPA-enhanced MRI with HBP is shown in Fig. 4, with an area under the curve of 95% (95% CI: 93%, 97%). Summary estimates of the PLR and NLR for Gd-EOB-DTPA-enhanced MRI with HBP are shown in Fig. 5. Because true-negative data could not be extracted for several studies, the diagnostic accuracy of Gd-EOB-DTPA-enhanced MRI without HBP was not able to be summarized. The specificity of those studies in set 2 ranged from 81 to 100%.

Fig. 3
figure 3

Summary estimates for the detection of HCC by using Gd-EOB-DTPA-enhanced MRI with HBP

Full size image
Fig. 4
figure 4

Summary ROC curve of Gd-EOB-DTPA-enhanced MRI with HBP

Full size image
Fig. 5
figure 5

Summary estimates of the PLR and NLR for Gd-EOB-DTPA-enhanced MRI with HBP

Full size image

Sensitivity analysis

Data from set 1 and set 2 were collected for Gd-EOB-DTPA-enhanced MRI with and without HBP to analyze summary sensitivity, respectively. Gd-EOB-DTPA-enhanced MRI with HBP exhibited significantly higher pooled sensitivity than that without HBP (84% [95% CI: 77%, 90%] vs 68% [95% CI: 57%, 77%], p = 0.01). Significant heterogeneity was found for the sensitivity of both MRI with and without HBP (p < 0.001 for both) (Fig. 6).

Fig. 6
figure 6

Summary sensitivities for Gd-EOB-DTPA-enhanced MRI with and without HBP

Full size image

Subgroup analysis

Additional sensitivity estimations for different subgroups were performed for factors that potentially affect diagnostic sensitivity (Table 2).

Sensitivity was significantly lower for studies from Asia than studies from outside of Asia (83% vs 85% for MRI with HBP, 63% vs 77% for MRI without HBP). However, no statistically significant difference was found between this subgroup (p > 0.05). In our meta-analysis, a case-control study was defined as a study containing both patients with HCC and a control group of subjects with hepatic lesions other than HCC. A case series study was defined as a study only including patients with hepatocellular carcinoma. Sensitivity of case-control study versus case series study in MRI without HBP was 77% and 57%, with p = 0.03. For MRI with HBP, there is no statistically significant difference in this subgroup.

Publication bias

The Deeks funnel plot showed a symmetrical appearance. A low likelihood of publication bias was performed for MRI with HBP in diagnostic accuracy analysis (p = 0.62, Supporting Figure S2). Low likelihood of publication bias in the sensitivity analysis was also observed for MRI with HBP (p = 0.62, Supporting Figure S3) and MRI without HBP (p = 0.98, Supporting Figure S4).

Discussion

Our results showed that Gd-EOB-DTPA-enhanced MRI with HBP for diagnosis HCC demonstrated a high sensitivity of 87% and a specificity of 92%. In a recent meta-analysis, Kierans et al demonstrated the performance of dynamic contrast material-enhanced MRI in the diagnosis of small (≤ 2 cm) HCC based on current guidelines as arterial enhancement and washout [54]. In that meta-analysis, a pooled sensitivity of 87% for MRI with HBP was reported in subgroup analysis, which was higher than those without HBP (pooled sensitivity of 65%). However, the imaging diagnostic criteria of their study were not unified due to some original research referring to DWI in diagnosing HCC. Our meta-analysis clearly defined the diagnostic algorithm and calculated relevant data from the original studies. Compared with extracellular contrast agent MRI, lower depiction of arterial hyperintensity due to the smaller amount of gadolinium injected and problem of pseudowashout for hepatic nodules may appear on Gd-EOB-DPTA MRI. But in our study, extending washout appearance to the transitional or hepatobiliary phase reached a favorable sensitivity. The diagnostic algorithm for Gd-EOB-DTPA-enhanced MRI with HBP had a significantly higher sensitivity than enhanced MRI without HBP (84% vs 68%, p = 0.01) for diagnosis of HCC. The summarized sensitivity of MRI with HBP images for evaluating HCC in any size was 84% in the meta-analysis by Lee et al [55]. In this meta-analysis, they did not evaluate the sensitivity of MRI with HBP images by lesion size or indicate the diagnostic algorithm in the subgroup. Lesion size is an important index in surveillance, diagnostic algorithm, and decisions regarding liver transplantation for hepatocellular carcinoma [7, 56].

Nonhepatocellular tumors also appear hypointense on HBP due to lacking of the organic anion transporting polypeptide, which may reduce the specificity for diagnosing HCC. Excluding benign lesions or non-HCC malignancies will maintain a relatively similar specificity for Gd-EOB-DTPA-enhanced MRI. Our meta-analysis reached a pooled specificity of 92%. The explanation may be that non-HCC malignancies are less frequently encountered in patients with chronic liver disease. Consistent with our meta-analysis, many studies have demonstrated that the addition of HBP in Gd-EOB-DTPA-enhanced MRI allowed higher sensitivity and without a significant reduction in specificity for diagnosis on HCC in patients with chronic liver disease [19, 21, 22, 29, 35, 37, 57, 58].

Subgroup analysis showed that case-control studies demonstrated higher sensitivity than case series studies (77% versus 57%, p = 0.03). A possible explanation for these results is that most of the case series studies were small sample size and included lesions smaller than 1 cm only. The classification of patients using Child-Pugh may be another potential factor contributing to heterogeneity. However, it was insufficient to obtain relative information from original studies, and the sample was too small for subgroup analysis. Large sample and high-quality studies are needed to verify this assumption. It is crucial that future studies adopt study designs that can better control biases and provide higher levels of evidence such as cohort studies and randomized controlled trials.

Our meta-analysis has several limitations. First, all the included studies were significantly heterogeneous, which affected the applicability of the summary estimate for diagnosis performance. However, this heterogeneity was useful for the subgroup analysis and may be a factor associated with improving sensitivity of MRI for detecting HCC. Besides, we used random-effects model and the summary ROC model to overcome the heterogeneity. The 95% CIs of sensitivity and specificity were not substantially wide, which demonstrates these results are valuable. Second, only 7 prospective studies were included in our analysis. This may cause a methodological limitation of including a relatively large number of retrospective studies. Pooling such suboptimal retrospective results may have caused a bias toward increased diagnostic sensitivity [59]. Third, we only analyzed the hepatobiliary agent of Gd-EOB-DTPA. Diagnosis performance of other hepatobiliary agents needs to be further analyzed in the future. Finally, our meta-analysis only included studies published in English, which possibly leads to “Tower of Babel” bias [60]. This bias refers to the possible tendency of investigators speaking non-English to only publish studies with positive results in international journals.

In conclusion, the results of our meta-analysis suggest that Gd-EOB-DTPA-enhanced MRI with HBP images has significantly higher sensitivity and comparable specificity than enhanced MRI without HBP images for diagnosis of HCC measuring up to 3 cm in patients with chronic liver disease. Therefore, the addition of HBP images in Gd-EOB-DTPA-enhanced MRI should be regarded as a major imaging feature for diagnosis HCC measuring up to 3 cm in patients with chronic liver disease.