Introduction

According to the current guidelines on the management of hepatocellular carcinoma (HCC), for hepatic nodules larger than 1 cm in patients with cirrhosis or chronic liver disease (CLD), HCC can be diagnosed without requirement of biopsy if the typical dynamic enhancement pattern, i.e., hypervascularity on the hepatic arterial phase (HAP) and washout on a later phase, is present in dynamic contrast-enhanced CT and/or MRI [1–5]. Recently, the accumulating evidence appears to suggest that MRI using hepatobiliary agents such as gadoxetic acid may be a more sensitive method for the detection of small HCCs than extracellular contrast media (ECCM)-enhanced CT or MRI [6–9]. This higher sensitivity is mainly attributable to the dual properties of gadoxetic acid which provides vascular phase information as well as additional information regarding the hepatobiliary phase (HBP) by its uptake into hepatocytes [10]. However, at present, the major guidelines proposed by the American Association for the Study of Liver Diseases (AASLD) in 2010 and the European Association for the Study of the Liver (EASL) in 2012 include ECCM-enhanced CT and MRI, not gadoxetic acid-enhanced MRI, as the standard methods for evaluating the vascular pattern of hepatic nodules [3, 11].

Despite the fact that the HBP of gadoxetic acid-enhanced MRI has been reported to facilitate the detection of small HCCs [9, 12–14] and to help differentiate HCCs from hypervascular pseudolesions [15, 16], hypointensity on the HBP or transitional phase (TP, 3-min delay) is a major problem in the interpretation of cirrhotic nodules based on their enhancement pattern [17]. This is because hypointensity on the TP and HBP of gadoxetic acid-enhanced MRI has a different meaning from the hypointensity observed on ECCM-enhanced imaging [18, 19]. For example, as the uptake of gadoxetic acid by hepatocytes starts approximately 90 s after contrast injection, the tumor-to-liver contrast in the TP would be derived not only from true washout of the contrast agents, but also from prominent enhancement of the surrounding hepatic parenchyma [18, 19]. Therefore, determining whether or not the hypointensity observed on dynamic phases and/or HBP can be used as an alternative to “washout”, or in other words, which phase(s) should be included to assess “washout,” is warranted for the application of gadoxetic acid-enhanced MRI for the noninvasive diagnosis of HCC.

Therefore, the purposes of our study are to evaluate the diagnostic performance of gadoxetic acid-enhanced MRI in the noninvasive diagnosis of HCC based on current guidelines of “hypervascularity on the HAP and washout on a later phase” and to determine the most appropriate dynamic phase(s) of gadoxetic acid-enhanced MRI to assess “washout” in a retrospective cohort of patients with CLD and arterially enhancing hepatic nodules.

Materials and methods

Patients

This retrospective cohort study was approved by our institutional review board and the requirement for informed consent was waived. From September 2012 to May 2013, a total of 2,874 gadoxetic-acid enhanced MRI images were available on our picture archiving and communication system (PACS). Among them, we included patients proven to have CLD by positive serum hepatitis B and/or C viral markers, histopathology, or clinical diagnosis of liver cirrhosis [20], in whom arterially enhancing hepatic nodules measuring 1 cm or larger in diameter were detected during imaging surveillance and in whom the final diagnosis was confirmed as either HCCs or non-HCCs. Accordingly, we excluded 2,404 cases upon review of their MRI reports and the electronic medical records (EMR) of each case. Reasons for exclusion with the corresponding numbers of patients are listed in Fig. 1. The remaining 470 scans were reviewed in consensus by two experienced radiologists (J.M.L. and I.J. with 22 and 7 years of experience in abdominal MRI). After image review of the 470 MRI scans, 182 cases which did not satisfy the inclusion criteria were additionally excluded (Fig. 1). The presence of arterial hyperenhancement was determined by visual comparison of the HAP and the unenhanced phase with additional review of subtraction images [21]. Nodules exhibiting homogeneous or variegated enhancement patterns were included while those exhibiting the ring enhancement pattern or peripheral globular enhancement pattern on the HAP were excluded [22]. Of the 288 patients enrolled in the final cohort, 227 were men and 61 were women with a mean age of 59 years (range, 21 – 85 years). Their chronic liver disease was associated with HBV (n = 188), HCV (n = 26), alcohol (n = 34), mixed causes (e.g., HBV and alcohol) (n = 6), and other reasons such as autoimmune hepatitis or unknown causes (n = 34). According to the Child-Pugh classification [23], 271 patients were classified as having Child-Pugh A, eight as Child-Pugh B, and three as Child-Pugh C, and the other six were unclassified due to the lack of laboratory findings. In cases with multiple lesions satisfying the inclusion criteria, the two aforementioned radiologists (J.M.L. and I.J.) selected and annotated the target lesions in the HAP images of gadoxetic acid-enhanced MRI for a maximum of three in number per patient. Finally, a total of 387 nodules in 288 MRI scans of 288 patients were included for imaging analysis.

Fig. 1
figure 1

Flowchart showing the study population and excluded patients

Full size image

Lesion confirmation

Of the 387 nodules, 292 nodules were diagnosed as HCCs (mean size, 35 mm; range, 10 – 210 mm), and the other 95 nodules were determined to be non-HCCs (mean size, 20 mm; range, 10 – 125 mm). Diagnoses of the 292 HCCs were made through histopathology (n = 128: surgical resection, n = 115; and percutaneous biopsy, n = 13), or by the typical enhancement pattern of HCCs according to the 2010 AASLD guidelines on dynamic liver CTs performed within 2 months prior to the MRI, i.e., arterial hyperenhancement and portal-venous or delayed phase washout (n = 164) [2].

Of the 95 non-HCCs, all 15 malignant lesions were confirmed by histopathology (n = 15: surgical resection, n = 8; percutaneous biopsy, n = 7). They included combined hepatocellular and cholangiocarcinomas (cHCC-CCs) (n = 3), intrahepatic cholangiocarcinomas (n = 10), and metastases (n = 2). Of the 80 non-HCC benign nodules, benignity was confirmed in nine by histopathology (surgical resection, n = 7; and percutaneous biopsy, n = 2) or by characteristic imaging findings and follow-up of a minimum of 6 months (n = 71). Of the nine benign lesions confirmed by histopathology, one nodule was diagnosed as an inflammatory lesion, seven were diagnosed as focal nodular hyperplasias (FNH) or FNH-like nodules based on their pathologic report, and one nodule was diagnosed as an arterioportal shunt (APS) as it was not found in the histology specimen and showed typical imaging features of APS on preoperative imaging studies described later in this paragraph. The imaging findings used for the imaging-based diagnosis of the 71 non-HCC benign lesions were as follows: for hemangioma (n = 34), its classic enhancement pattern on dynamic contrast-enhanced CT or MRI and hyperintensity on T2-weighted MR imaging, without interval change during follow-up [24, 25]; for APS (n = 32), isoattenuation or isointensity on dynamic CT or MRI other than HAP images, isointensity on the HBP, and no change or a decrease in size during follow-up [15, 26]; for the inflammatory lesion (n = 1), conspicuity on at least one of either the portal-venous or delayed phase images different from APS, and a decrease in size or disappearance during the follow-up period [27, 28]; and for FNH or FNH-like nodules (n = 4), iso- or hypersignal intensity during the portal-venous phase, at least partial hyperintensity compared to the liver parenchyma in the hepatobiliary phase, and no interval change [29]. In summary, the final diagnosis of non-HCC lesions were 15 malignant lesions including cHCC-CCs (n = 3), intrahepatic cholangiocarcinomas (n = 10), metastases (n = 2), and 80 benign lesions including hemangiomas (n = 34), APS (n = 33), inflammatory lesions (n = 2), and FNH or FNH-like nodules (n = 11). Subgroup analyses were additionally performed for lesions diagnosed through histopathology (128 HCCs and 24 non-HCCs) in order to avoid potential false positive and false negative diagnoses from imaging-based diagnosis, and for lesions smaller than 20 mm in diameter (97 HCCs and 70 non-HCCs).

MR image acquisition

MR examinations were performed on either the 3.0 T or 1.5 T MR system available in our institution: Signa HDxt 1.5 T (GE Medical Systems, Milwaukee, WI, USA) (n = 127), Magnetom Verio or Magnetom Trio or Biograph mMR (Siemens Healthcare, Erlangen, Germany) (n = 53), Ingenia 3.0 T (Philips Medical System, Best, Netherlands) (n = 53), or various 3.0 T or 1.5 T scanners at an outside hospital (n = 55). Routine MRI sequences included a respiratory-triggered T2-weighted fast spin echo sequence, a half-Fourier acquisition single-shot turbo spin-echo sequence (HASTE), free-breathing diffusion-weighted imaging (DWI), and a breath-hold T1-weighted spoiled dual echo (in-phase and out-of-phase) gradient recalled echo (GRE) sequence. Dynamic and HBP images were obtained using the fat-suppressed, 3D GRE sequence. Detailed parameters of the T1-weighted 3D GRE sequence are listed in Table 1. Immediately after obtaining the unenhanced T1-weighted images, 10 mL of gadoxetic acid (Primovist, Bayer Schering Pharma AG, Berlin, Germany) was administered intravenously at a rate of 1.5 mL/s, followed by a 20 mL saline flush through an antecubital venous catheter, and thereafter, multiphasic dynamic images were obtained. Scanning delay times were determined with real-time MRI fluoroscopic monitoring. Arterial phase images were obtained 7 s after contrast arrival at the distal thoracic aorta, and subsequent portal-venous phase (PVP) images were obtained approximately 50 s after beginning contrast medium injection. Thereafter, TP and HBP images were obtained 3 min and 20 min after beginning contrast medium injection, respectively. Acquisition of 3D GRE data for each dynamic phase and HBP was finished during a single breath-hold (15 – 20 s) at the end of expiration.

Table 1 MR Acquisition parameters of the T1-weighted 3D GRE sequence at our institution
Full size table

Image analysis

All MR images were interpreted in consensus by two abdominal radiologists (D.H.L. and J.H.J., each with 8 years of experience in the interpretation of liver MR imaging) who were blinded to whether the lesions were confirmed as HCCs or non-HCCs. For each target lesion, which was annotated on the HAP images, two radiologists evaluated whether or not the lesion showed hypointensity on the PVP, TP, and HBP images, respectively. Hypointensity on the PVP, TP, and HBP were qualitatively defined as when any part of the nodule showing arterial enhancement demonstrated a corresponding hypointense area relative to the surrounding liver parenchyma (within 3 cm from the outer border of the nodule) [30].

Statistical analysis

Based on the results of imaging interpretation, three different gadoxetic acid-enhanced MR imaging criteria for “washout on a later phase” were applied for the diagnosis of HCCs. Those MR imaging criteria were: hypervascularity on the HAP and (1) hypointensity on the PVP, (2) hypointensity on the PVP and/or TP, or (3) hypointensity on the PVP and/or TP, and/or HBP (i.e. hypointense on one or more of later phases). For each criterion, per-lesion sensitivity, specificity, and positive and negative predictive values with 95 % confidence intervals (95 % CIs), and accuracies were calculated. All statistical analyses were performed using MedCalc software version 12.4.0.0 (MedCalc Software, Mariakerke, Belgium).

Results

Frequency of hypointensity during the PVP, TP, and HBP

Of the 387 arterially enhancing nodules (292 HCCs and 95 non-HCCs) on gadoxetic acid-enhanced MRI in patients with CLD, 209 nodules (207 HCCs and two non-HCCs) (54.0 %) showed relative hypointensity compared with the surrounding parenchyma on the PVP; 264 nodules (251 HCCs and 13 non-HCCs) (68.2 %) showed hypointensity on the TP; and 318 nodules (269 HCCs and 49 non-HCCs) (82.2 %) were seen as hypointense on the HBP. Among small nodules < 20 mm in diameter (n = 167), 54 nodules (54 HCCs and 0 non-HCCs) (32.3 %) showed hypointensity on the PVP; 84 nodules (78 HCCs and six non-HCCs) (50.3 %) showed hypointensity on the PVP and/or TP; and 117 nodules (87 HCCs and 30 non-HCCs) (70.1 %) showed hypointensity on the PVP and/or TP, and/or HBP.

Diagnostic performance of gadoxetic acid-enhanced MRI using different criteria for “washout”

For the differential diagnosis of HCCs from other arterially enhancing nodules on gadoxetic acid-enhanced MRI, diagnostic performances of the different imaging criteria for “washout” are summarized in Table 2. Imaging criterion of “hypointensity on the PVP” provided the highest specificity (97.9 %) and highest positive predictive value (PPV) (99.0 %) for the diagnosis of HCCs, whereas criteria of “hypointensity on the PVP and/or TP” and “hypointensity on the PVP and/or TP, and/or HBP” showed lower specificities and lower PPVs of 86.3 % and 95.1 %, and 48.4 % and 84.8 %, respectively (Fig. 2).

Table 2 Diagnostic Performances of Different Imaging Criteria for “Washout” on Gadoxetic Acid-Enhanced MRI for Hepatocellular Carcinoma
Full size table
Fig. 2
figure 2

A histopathologically confirmed hepatocellular carcinoma (HCC) in a 31-year-old male patient with chronic hepatitis B. On gadoxetic acid-enhanced MRI, there is an arterially enhancing nodule in the caudate lobe of the liver (a) showing hypointensity relative to the liver parenchyma in the portal-venous phase (PVP) (b) and transitional phase (TP) images (c) (arrows). (d) On the hepatobiliary phase (HBP) image, the nodular lesion shows hypointensity relative to the surrounding parenchyma. All imaging criteria for washout including “hypointensity on the PVP”, “hypointensity on the PVP and/or TP”, and “hypointensity on the PVP and/or TP, and/or HBP” can correctly diagnose this nodule as an HCC

Full size image

Of the 95 non-HCCs, two lesions which showed “hypointensity on the PVP” were cHCC-CCs, while 13 lesions which showed “hypointensity on the PVP and/or TP” included cHCC-CCs (n = 3) (Fig. 3), cholangiocarcinomas (n = 6), metastases (n = 1), and hemangiomas (n = 3) (Fig. 4). Of the 292 HCCs, 274 HCCs were seen as hypointense lesions on one or more of the later phases including the PVP, TP, and HBP resulting in the highest sensitivity of 93.8 %. However, criterion of “hypointensity on the PVP and/or TP, and/or HBP” demonstrated the lowest specificity of 48.4 % as many of the non-HCCs (n = 49) were seen as hypointense lesions on HBP: cHCC-CCs (n = 3), cholangiocarcinomas (n = 10), metastases (n = 2), hemangiomas (n = 32), and inflammatory lesions (n = 2).

Fig. 3
figure 3

A histopathologically confirmed combined hepatocellular and cholangiocarcinoma in a 60-year-old female patient with chronic hepatitis B. (a) On hepatic arterial phase image, a 4 cm tumour with variegated enhancement (arrow) is seen in segment 6 of the liver. On PVP (b), TP (c), and HBP (d), the arterially enhancing portion of this tumour shows iso-, hypo-, and hypointensity relative to the liver parenchyma, respectively. Criterion of “hypointensity on the PVP” would correctly diagnose this lesion as non-HCC; however, criteria of “hypointensity on the PVP and/or TP” and “hypointensity on the PVP and/or TP, and/or HBP” would lead to incorrect diagnoses of this lesion as an HCC

Full size image
Fig. 4
figure 4

A hemangioma in a 61-year-old male patient with chronic hepatitis B. (a) A small arterially enhancing nodule (arrow) is seen in the subcapsular area of the left lobe of the liver. This lesion shows isointensity relative to the liver parenchyma in the PVP images (b) and hypointensity in the TP (c) and HBP images (d). Imaging criterion of “arterial hyperenhancement and hypointensity on the PVP” would correctly diagnose this lesion as non-HCC; however, criteria of “hypointensity on the PVP and/or TP” and “hypointensity on the PVP and/or TP, and/or HBP” would lead to incorrect diagnoses of this lesion as an HCC. This lesion showed hyperintensity on the T2-weighted image (e), and 6-month follow-up MR image showed no interval change (f)

Full size image

In the subgroup analysis of histopathologically confirmed lesions (128 HCCs and 24 non-HCCs), diagnostic performances of the different imaging criteria for “washout” are summarized in Table 3. In small nodules < 20 mm in diameter (97 HCCs and 70 non-HCCs), imaging criterion of “hypointensity on the PVP” resulted in a sensitivity of 55.7 % (54/97) and a specificity of 100 % (70/70), whereas imaging criteria of “hypointensity on the PVP and/or TP” and “hypointensity on the PVP and/or TP, and/or HBP” resulted in sensitivities of 80.4 % (78/97) and 89.7 % (87/97), and specificities of 91.4 % (64/70) and 57.1 % (40/70), respectively. Subgroup analyses revealed similar trends of sensitivity, specificity, and accuracy with the analysis of the whole study group.

Table 3 Diagnostic Performances of Different Imaging Criteria for “Washout” on Gadoxetic Acid-Enhanced MRI for Hepatocellular Carcinoma: Subgroup Analysis of Histopathologically Confirmed Lesions
Full size table

Discussion

For the noninvasive diagnosis of HCC based on “hypervascularity in the HAP and washout on a later phase”, our retrospective cohort study demonstrated that the criterion of “hypointensity on the PVP” of gadoxetic acid-enhanced MRI resulted in a high specificity of 97.9 %, while the criteria of “hypointensity on the PVP and/or TP” and “hypointensity on the PVP and/or TP, and/or HBP” resulted in lower specificities of 86.3 % and 48.4 %, respectively. Considering the clinical importance of obtaining a specific diagnosis of HCC, these results suggest that washout should be assessed in the PVP alone rather than combined with the TP or HBP for a more specific diagnosis of HCCs on gadoxetic acid-enhanced MRI [31]. Obtaining a specific diagnosis of HCC, i.e. minimizing false positives, is of critical importance in clinical practice because a false positive diagnosis of HCC can result in inappropriate or unnecessary treatment in patients with non-HCC malignant lesions or benign lesions, as well as unfair priority granting for liver transplantations [32]. Thus, a stringent diagnostic criterion is required for the noninvasive diagnosis of HCCs [32]. This is also why contrast enhanced ultrasound which has been reported to cause false positive HCC diagnoses in patients with cholangiocarcinoma has been eliminated from the 2010 AASLD guidelines [33]. As the current guidelines based on the typical vascular pattern has been reported to allow a correct diagnosis of HCCs with a predicted specificity of 95 % using ECCM-enhanced imaging [34, 35], our results of gadoxetic acid-enhanced MRI showing a specificity of 97.9 % based on the criterion of “hypointensity on the PVP” suggest that gadoxetic acid-enhanced MRI can also be used to evaluate the vascular pattern of hepatic nodules for the diagnosis of HCC. This specificity is also comparable to the specificities (96 – 97 %) of ECCM-enhanced CT or MRI [36, 37].

As for the sensitivity of gadoxetic acid-enhanced MRI in the diagnosis of HCCs, our study showed that relative hypointensity of the tumour was present in 70.9 % on the PVP and 86.0 % on the TP, which is comparable to a recent study on ECCM-enhanced MRI showing sensitivities of 44 % with PVP and 82 % with the delayed phase [38]. In addition, 93.8 % of HCCs in our study were seen as hypointense lesions on the HBP. Owing to the high percentage of HCCs showing hypointensity on the HBP, the criterion of “hypointensity on the PVP and/or TP, and/or HBP” resulted in high sensitivity. However, that criterion also resulted in low specificity as many non-HCC lesions which lacked normal hepatocytes would show hypointensity on the HBP. Therefore, we conclude that the HBP should not be used as a phase for assessing “washout” owing to its low specificity.

In our study, 13 non-HCCs showed hypointensity relative to the liver parenchyma on the TP. This resulted in a number of false positives when applying the criterion of “hypointensity on the PVP and/or TP.” As hemangioma is the most common benign hepatic neoplasm and usually does not require treatment, accurate imaging diagnosis of hemangioma is of importance [18]. While hemangiomas typically show delayed retention of contrast media on ECCM-enhancing imaging [39], our study demonstrated that three hemangiomas were seen to be hypointense on the TP of gadoxetic acid-enhanced MRI. This result is in good agreement with previous studies by Doo et al. [19] which reported that high-flow hemangioma may show the “pseudo-washout” sign during the TP. Tateyama et al. [18] also reported that prolonged enhancement of hepatic hemangioma was less frequent on gadoxetic-acid enhanced MRI (52 %) than contrast-enhanced CT (100 %). This observed hypointensity of hemangioma on the TP can be due to the uptake of gadoxetic acid by the surrounding liver parenchyma, the lower dose of gadolinium, and the shorter plasma half-life of gadoxetic acid [40]. Interestingly, hypointensity on the PVP and/or TP of high-flow hemangioma has also been reported on gadobenate dimeglumine (another hepatocyte-specific contrast agent)-enhanced MRI [41]. There were other false positives on the TP in addition to hemangioma including malignant lesions such as cHCC-CCs which will be described later, cholangiocarcinomas, and metastases. As for intrahepatic cholangiocarcinomas in the cirrhotic liver, it would be more frequently hypervascular than those in the normal liver [42], and thus is already known as a common mimicker of HCCs. Rimola et al. [43] suggested that the absence of washout of cholangiocarcinomas in the cirrhotic livers on the delayed phase of gadodiamide (an ECCM)-enhanced MRI may help avoid a potential misdiagnosis. However, as our results demonstrated, hypervascular intrahepatic cholangiocarcinomas can show hypointensity in the TP of gadoxetic acid-enhanced MRI. Peporte et al. [44] also reported that cholangiocarcinoma can be seen as a hypointense nodule on the TP.

Finally, we found that using the criterion of “hypointensity on the PVP”, two non-HCCs were diagnosed as HCCs. These two false positive cases in our study were confirmed as cHCC-CCs. None of the benign lesions showed hypointensity on the PVP. As cHCC-CC is histologically composed of elements from both entities, its vascular pattern on contrast-enhanced imaging depends on the proportion of the tumour components [45, 46]. Therefore, an HCC-dominant tumour can show arterial hypervascularity and washout on a later phase mimicking an HCC, leading to a false positive diagnosis even with CT or ECCM-enhanced MRI [47, 48]. However, as patients with cHCC-CC have been reported to respond differently to therapy and that the prognoses after liver transplantation are poorer than those with HCC [49–51], differentiating cHCC from HCC would be very important in the establishing noninvasive diagnostic criteria. Therefore, further studies are warranted to differentiate cHCC-CCs from HCCs using imaging features.

This study has several limitations. First, as our study was performed retrospectively, there may have been selection bias in our patient population. However, we put our best effort to collect study patients in a consecutive cohort in order to avoid or minimize selection bias. Second, histological diagnoses were not available for a substantial number of benign non-HCCs. As performing biopsy for non-HCC lesions showing characteristic imaging features of common benign lesions may be unethical, characteristic imaging findings and follow-ups with CT or MRI were used for confirmation of the diagnosis. Third, as not all patients underwent ECCM-enhanced CT or MRI and gadoxetic acid-enhanced MRI within a short interval, we were not able to compare the diagnostic performances of the two imaging techniques and thus further study would be warranted in this regard. Fourth, we solely focused on the hemodynamic pattern of gadoxetic acid-enhanced MRI, and did not assess the role of ancillary features. Considering that static MRI sequences such as T2-weighted imaging and DWI with apparent diffusion coefficient (ADC) mapping can provide additional information other than just the hemodynamic pattern, the performance of MRI in diagnosing HCC using criteria including these ancillary features may be different from our study results.

In conclusion, when gadoxetic acid-enhanced MRI is used for the noninvasive diagnosis of HCC based on “hypervascularity on the HAP and washout on a later phase,” washout should be determined on the PVP alone rather than combined with hypointensity on the TP or HBP in order to maintain high specificity.