Hepatocellular adenomas: is there additional value in using Gd-EOB-enhanced MRI for subtype differentiation?

Purpose To differentiate subtypes of hepatocellular adenoma (HCA) based on enhancement characteristics in gadoxetic acid (Gd-EOB) magnetic resonance imaging (MRI). Materials and methods Forty-eight patients with 79 histopathologically proven HCAs who underwent Gd-EOB-enhanced MRI were enrolled (standard of reference: surgical resection). Two blinded radiologists performed quantitative measurements (lesion-to-liver enhancement) and evaluated qualitative imaging features. Inter-reader variability was tested. Advanced texture analysis was used to evaluate lesion heterogeneity three-dimensionally. Results Overall, there were 19 (24%) hepatocyte nuclear factor (HNF)-1a-mutated (HHCAs), 37 (47%) inflammatory (IHCAs), 5 (6.5%) b-catenin-activated (bHCA), and 18 (22.5%) unclassified (UHCAs) adenomas. In the hepatobiliary phase (HBP), 49.5% (39/79) of all adenomas were rated as hypointense and 50.5% (40/79) as significantly enhancing (defined as > 25% intralesional GD-EOB uptake). 82.5% (33/40) of significantly enhancing adenomas were IHCAs, while only 4% (1/40) were in the HHCA subgroup (p < 0.001). When Gd-EOB uptake behavior was considered in conjunction with established MRI features (binary regression model), the area under the curve (AUC) increased from 0.785 to 0.953 for differentiation of IHCA (atoll sign + hyperintensity), from 0.859 to 0.903 for bHCA (scar + hyperintensity), and from 0.899 to 0.957 for HHCA (steatosis + hypointensity). Three-dimensional region of interest (3D ROI) analysis showed significantly increased voxel heterogeneity for IHCAs (p = 0.038). Conclusion Gd-EOB MRI is of added value for subtype differentiation of HCAs and reliably identifies the typical heterogeneous HBP uptake of IHCAs. Diagnostic accuracy can be improved significantly by the combined analysis of established morphologic MR appearances and intralesional Gd-EOB uptake. Key Points • Gd-EOB-enhanced MRI is of added value for subtype differentiation of HCA. • IHCA and HHCA can be identified reliably based on their typical Gd-EOB uptake patterns, and accuracy increases significantly when additionally taking established MR appearances into account. • The small numbers of bHCAs and UHCAs remain the source of diagnostic uncertainty.

This new classification has led to a change in treatment algorithms [4,12,13].
In the updated molecular classification of 2017, b-cateninactivated adenomas are furthermore subdivided into an exon 3-mutated and an exon 7-8-mutated subtype. Another subtype has been identified recently and has been termed sonic hedgehog HCA (shHCA) based on its molecular pathway. Previously counted among unclassified adenomas, shHCAs are assumed to account for about 5% of all adenomas [14,15]. Generally, lesions > 5 cm should be resected because of their increased risk of rupture, bleeding, and malignant transformation. For lesions < 5 cm, treatment is more individually based on the histological subtype. IHCAs have been associated with the presence of hepatic steatosis [16].
Following precise diagnostic characterization, an individual estimate of a patient's risks and possible complications has to be performed. As HHCAs and small IHCAs are less likely to transform into hepatocellular carcinomas (HCCs), they are typically managed by regular follow-up at intervals of 6-12 months [17]. In general, patients with diagnostically proven HCA should avoid any triggers adenomas are linked with, for instance by intermitting oral contraceptives (OCPs) or steroids and/or by lifestyle modification to lower their body mass index (BMI) [3,12,18]. Resection or close follow-up is indicated in patients with bHCA, which is believed to be more prone to malignant transformation and is more often observed in men or patients with, or concomitant glycogen storage disease [15,16]. Overall, malignant transformation is relatively rare, occurring in 5-10% of all patients, and is most commonly observed in bHCA [15,16]. Reports in the literature describe bHCA/IHCA mixed types, which tend to be associated with a higher risk of malignancy. In addition, IHCAs in general are associated with a higher risk of bleeding; therefore, patients with these adenomas should also be monitored more closely [8,15,16,19,20].
Several recent studies have shown the value of magnetic resonance imaging (MRI) with use of a liver-specific contrast agent such as gadobenate dimeglumine (Gd-BOPTA; MultiHance, Bracco Imaging) or gadoxetic acid (Gd-EOB; Primovist or Eovist, Bayer Pharma) in differentiating focal liver lesions (especially FNH and HCA) [19,[21][22][23][24][25][26]. Gd-EOB seems to be costeffective for differentiation of malignant and benign liver lesions in routine clinical management [27] and might also have the potential to differentiate HCA subtypes. However, only a few studies have addressed this issue so far. Reliable noninvasive identification of (even small) IHCA or bHCA could change clinical management because of the increased risk of bleeding or malignant transformation of these histological subtypes [8,16,20]. Published data suggest that, on hepatobiliary phase (HBP) images, most HCAs are hypointense compared with surrounding liver parenchyma, while a minority of HCAs are iso-or even hyperintense [21,23,28]. Conversely, more IHCAs and bHCAs appear to be iso-or hyperintense on HBP images, while HHCAs have been found to have the lowest late-phase intensity [28][29][30].
The purpose of this study was to investigate the diagnostic value of Gd-EOB-enhanced MRI in differentiating histological subtypes of HCAs.

Patients
Our institutional review board approved this retrospective study (internal registration number EA2/016/14) and waived informed consent due to the retrospective nature. The study protocol conforms to the ethical guidelines of the 2002 Declaration of Helsinki. All patients with histopathologically confirmed HCA who underwent Gd-EOB-enhanced liver MRI between January 2009 and March 2019 were retrospectively identified from the institutional databases.
Only surgically resected lesions were included. A "blinded" pathologist evaluated all macroscopic and microscopic features, which was followed by reprocessing for immunohistochemical analysis. HCAs were classified into the four major molecular subgroups according to their genetic and phenotypic characteristics (HNF-1a-mutated adenoma (HHCA), inflammatory adenoma (IHCA), β-catenin-activated adenoma (bHCA), and unclassified adenoma (UHCA)) [4,8,12,13]. Some of the patients included in the present analysis participated in a previous HCA study that did not include a subgroup analysis [23]. We identified and enrolled a total of 48 patients with 79 HCA lesions. There were 44 female and four male patients. They had a mean age of 38.5 ± 10.5 years (range, 20-67 years).

Quantitative analysis
Intralesional Gd-EOB heterogeneity All lesions were segmented with 3D regions of interest (ROIs) to encompass the whole tumor volume in the HBP using dedicated segmentation software (Medical Imaging Toolkit, MITK). After segmentation, voxel-based texture analysis was performed regarding lesion heterogeneity using the pyradiomics algorithm.

Dynamic CE behavior
Polygonal 2D ROIs including the entire tumor at its maximum cross-sectional diameter were placed manually. Cystic and/or hemorrhagic components were spared if reasonably possible. All ROIs were placed in the arterial phase sequence and cloned to the subsequent contrast phase sequences (portal venous and transitional phases). An additional circular ROI with a fixed diameter of 10 mm was placed in healthy liver parenchyma not including vessels and bile ducts. Furthermore, lesion-to-liver enhancement was calculated for the different contrast phases as follows: Diagnostic accuracy in terms of sensitivity and specificity was determined by receiver operating characteristic (ROC) curve analysis. All tests were performed two-sided with a level of significance of 0.05. Boxplots were created with GraphPad Prism (GraphPad Software, Inc.). Voxel heterogeneity was evaluated using texture analysis (pyradiomics).

Quantitative analysis
Intralesional Gd-EOB heterogeneity Volumetric segmentation in the HBP with subsequent texture analysis was performed in 34 patients. Fourteen patients had to be excluded. To strengthen our subjective results of intralesional Gd-EOB heterogeneity in IHCA, we divided the population into "IHCA" and "non-IHCA". Volumebased analysis revealed significantly increased voxel heterogeneity for the IHCA group (variance of mean SI, 6465.48) compared with the non-IHCA group (variance of mean signal intensity (SI), 2861.80) (p = 0.038).

Dynamic CE behavior
ROI-based relative lesion-to-liver enhancement in the late phase was not found to differ significantly between subgroups (p = 0.490  (Fig. 4).

Discussion
Our results show that Gd-EOB MRI provides additional information for differentiating HCA subtypes. IHCAs can be differentiated reliably based on their typical heterogeneous HBP uptake behavior defined in our study as significant enhancement. Furthermore, when Gd-EOB uptake behavior is In a meta-analysis published in 2017, Guo et al described a low signal intensity (SI) for HCAs in the HBP and concluded that combining SI in the HBP with established MRI features and risk factors of liver disease might improve diagnostic yield in the detection and differentiation of HCAs. Guo et al further considered the benefit of Gd-EOB-enhanced imaging to be overrated [31].
The established MRI characteristics known so far were also present in our study. For instance, HHCAs showed a signal drop in the unenhanced opposed-phase series (95%; p < 0.001) consistent with diffuse intralesional steatosis. This is in line with Aalten et al (78%), Laumonier et al (87%), and Tse et al (100%), underlining the accuracy of this simple visual diagnostic sign (28,32,33). In 54% of IHCA lesions in our study, the atoll sign was present, yielding nearly 100% specificity for this feature as only one other adenoma also showed the atoll sign (Figs. 2 and 3 and Table 1 [32,33]. In our study population, we noticed a group of adenomas with heterogeneous Gd-EOB uptake in the HBP. Accordingly, we devised a 5-point scale system to subjectively estimate the degree of heterogeneity/level of hyperintensity in the late uptake phase. This was done by two readers blinded to the clinical data, and there was good inter-reader agreement. The enhancement behavior of HHCAs and IHCAs is unequivocal with HHCAs showing Fig. 2 Flowchart of HCA subgroup diagnostic algorithm including combined evaluation of Gd-EOB uptake behavior and established MRI features. Intralesional Gd-EOB uptake in the HBP was rated subjectively as iso-to hyperintensity percentage on a 5-point scale (score 0, 0%; score 1, 10-25%; score 2, 25-50%; score 3, 50-75%; score 4, > 75%) essentially no uptake and IHCAs appearing significantly more heterogeneous because of their uptake (p < 0.001) (Figs. 5 and 6). These subjective ratings are supported by our volume-based voxel analysis, which showed IHCAs to be significantly more heterogeneous than the other three subtypes.
In its Clinical practice guidelines on the management of benign liver tumours, the European Association for the Study of the Liver (EASL) also states that IHCA may present iso-or hyperintense in the HBP, emphasizing the results of a study by Ba-Ssalamah et al [16,30]. Ba-Ssalamah et al evaluated the degree of OATP1B1/3 and MRP3 expression, which they found to correlate statistically with Gd-EOB retention and washout in the HBP, resulting in 77% specificity for diagnosing IHCA [16,30]. In our study, we achieved 93% specificity when combining intralesional Gd-EOB heterogeneity with established MRI criteria. . c-f Contrast enhancement behavior in the T1w arterial phase with a mild hyperintense signal to the surrounding liver (c), an isointense signal in the portal venous phase (d) and mildly hypointense in the transitional phase (e). In the HBP (f), the lesion appears homogeneously hypointense without a significant uptake (0%) Fig. 4 Relative lesion-to-liver (%) enhancement in the arterial (art), portal venous (pv) and transitional (trans) phases for the four subgroups of HHCA, IHCA, bHCA, and UHCA. Kruskal-Wallis test showed significant differences for the arterial (p = 0.024) and portal venous (p = 0.018) phases and nonsignificant differences for the transitional phase (p > 0.05) The absence of HBP uptake and subjective impression of hypointensity we found for HHCAs are in line with the results of Tse et al, who found the lowest lesion-to-liver SI ratio for HHCAs and could thus distinguish the latter from other subtypes [28]. In our study cohort, we observed that HHCA and also UHCA can show homogeneous hypointensity in the HBP. Comparison of the two revealed that HHCA appeared more hypointense than UHCA. However, our T1w HBP was acquired with fat saturation so that the even lower signal of HHCAs can be explained by their higher amount of intralesional steatosis. Nevertheless, Tse et al were not able to differentiate more subtypes using Gd-EOB-enhanced imaging [28]. We improved diagnostic performance by categorizing lesions based on intralesional Gd-EOB uptake in the HBP and considering uptake behavior in conjunction with established MRI features. For IHCA (+atoll), bHCA (+scar), and HHCA (+fat), this combined approach resulted in specificities and sensitivities of at least 90%, which was further corroborated by AUC results (Figs. 1 and 2). Even if the results for bHCA lack validity because of the small number of lesions included in the analysis, the diagnostic accuracy reached here is promising (Figs. 1, 2 and 7). Like earlier investigators, we also encountered problems with the unclassified HCA subtype, which does not seem to fit into any pattern. . c-f Contrast enhancement behavior in the T1w arterial (hyperintense washin) (c), portal venous (heterogeneous washout) (d), and transitional (washout) (e) phases.
In both the portal venous (d) and venous phases, the adenoma appears mildly hypointense to the surrounding liver. In the HBP, the lesion shows heterogeneous patchy uptake behavior > 50% and is iso-to hyperintense to surrounding liver (white arrow-(f)) . c-f Contrast enhancement behavior in the T1w arterial phase with heterogeneous hyperintense washin (c), isointensity in the portal venous phase (d), and transitional signal intensity to the surrounding liver (e). In the HBP, the lesion shows heterogeneous uptake rated as low (0-25%) and is hypointense to the liver Besides Gd-EOB-enhanced imaging, a recently published study of Bise et al introduced two new MRI features with high diagnostic accuracy. For IHCAs, Bise et al described the isolated peripheral sinusoidal dilatation "as the crescent sign", defined as an incomplete hyperintense rim on T2w and/or arterial phase sequences [34]. For HHCAs, Bise et al described a hypovascular pattern; hence, this subtype is hypointense compared with surrounding liver on T1w FS sequences. The lesions with a hypovascular pattern were described as showing mild enhancement in the arterial phase and were found to be characteristically hypointense in the delayed phase [34]. As mentioned in the "Introduction", the new classification of HCA distinguishes six subtypes. Due to the retrospective character of both ours and Bise et al's study, MRI features have not yet been tested regarding the new molecular subtypes (bHCA subtypes mutated in exon 3 and 7-8 and shHCA subtype) [34]. Future studies need to evaluate and probably combine Bise et al's new MRI features and the specific Gd-EOB uptake behavior described by us regarding the new subtypes in order to characterize them noninvasively. Nault et al emphasized the clinical relevance of the new subtypes as bHCAs mutated in exon 3 are more likely to turn into HCC and shHCA are associated with frequent symptomatic bleeding [15]. Nevertheless, given that HCAs are altogether rare, it will take some time to identify enough cases for meaningful analysis.
There are several limitations of the present analysis. First, this is a retrospective study and patients were not enrolled consecutively. Second, as this analysis was restricted to histologically proven HCAs in order to exclude falsely classified lesions from analysis, a selection bias may have occurred. Third, our study is limited by the small number of bHCAs and the small sample size available for voxel-based volumetric analysis. Furthermore, although readers were blinded, they were aware of the study design, which may have introduced detection bias.
In conclusion, combining Gd-EOB uptake behavior in the hepatobiliary phase with established MRI criteria (atoll sign, intralesional fat, and central scar) leads to a high diagnostic accuracy in HCA subtype differentiation. Our approach may improve the noninvasive distinction of IHCA and HHCA and thus help in identifying those patients who might benefit from (early) resection to prevent bleeding or malignant transformation.

Compliance with ethical standards
Guarantor The scientific guarantor of this publication is Prof. Dr. med. Timm Denecke.
Conflict of interest Christian Grieser, Dominik Geisel, Daniel Seehofer, and Timm Denecke received honoraria, and travel expenses from Bayer Schering Pharma, Berlin, Germany in the past. All other authors who took part in this study declared that they do not have anything to disclose regarding funding or conflict of interest with respect to this manuscript.

Statistics and biometry
No complex statistical methods were necessary for this paper.
Informed consent Written informed consent was obtained from all subjects (patients) in this study.
Ethical approval Institutional review board approval of this retrospective study was obtained (No. EA2/016/14).
Study subjects or cohort overlap Some study subject or cohorts have been previously reported in: Grieser  Funding information Open Access funding enabled and organized by Projekt DEAL. The authors state that this work has not received any funding.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.