Contrast-enhanced computed tomography plus gadolinium-ethoxybenzyl diethylenetriamine pentaacetic acid-enhanced magnetic resonance imaging for gross classification of hepatocellular carcinoma

Accurate gross classification through imaging is critical for determination of hepatocellular carcinoma (HCC) patient prognoses and treatment strategies. The present retrospective study evaluated the utility of contrast-enhanced computed tomography (CE-CT) combined with gadolinium-ethoxybenzyl diethylenetriamine pentaacetic acid-enhanced magnetic resonance imaging (EOB-MRI) for diagnosis and classification of HCCs prior to surgery. Ninety-four surgically resected HCC nodules were classified as simple nodular (SN), SN with extranodular growth (SN-EG), confluent multinodular (CMN), or infiltrative (IF) types. SN-EG, CMN and IF samples were grouped as non-SN. The abilities of the two imaging modalities to differentiate non-SN from SN HCCs were assessed using the EOB-MRI hepatobiliary phase and CE-CT arterial, portal, and equilibrium phases. Areas under the ROC curves for non-SN diagnoses were 0.765 (95% confidence interval [CI]: 0.666–0.846) for CE-CT, 0.877 (95% CI: 0.793–0.936) for EOB-MRI, and 0.908 (95% CI: 0.830–0.958) for CE-CT plus EOB-MRI. Sensitivities, specificities, and accuracies with respect to identification of non-SN tumors of all sizes were 71.4%, 81.6%, and 75.5% for CE-CT; 96.4%, 78.9%, and 89.3% for EOB-MRI; and 98.2%, 84.2%, and 92.5% for CE-CT plus EOB-MRI. These results show that CE-CT combined with EOB-MRI offers a more accurate imaging evaluation for HCC gross classification than either modality alone.

Our previous studies showed that contrast-enhanced computed tomography (CE-CT) was a useful radiological technique for evaluating HCC gross classification [13]. Gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (Gd-EOB-DTPA) is a hepatocyte-specific magnetic resonance imaging (MRI) contrast agent with combined perfusion and hepatocyte-selective properties. In the hepatobiliary phase, HCCs without functioning hepatocytes usually show hypointensity compared with hyperintensity of the surrounding liver parenchyma owing to Gd-EOB-DTPA uptake by normal hepatocytes. This contrast agent reportedly enhances detection of small HCCs and predicts gross classification [14][15][16]. Contrast-enhanced MRI using Gd-EOB-DTPA (EOB-MRI) clearly distinguishes focal liver lesions from the surrounding hepatic parenchyma, and has the potential to precisely predict HCC gross classification [17].

ROC curve analyses
ROC curves for diagnosis of non-SN type HCC via CE-CT, EOB-MRI, and CE-CT plus EOB-MRI are shown in Figure 3. The AUC (95% confidence interval AUC values for CE-CT plus EOB-MRI, and EOB-MRI were higher than those for CE-CT for all tumor sizes (P = 0.006, 0.034, respectively) and HCCs ≥3 cm (P = 0.005, 0.016, respectively) ( Table 2). In HCCs > 3 cm, CE-CT plus EOB-MRI AUC values were also higher than those for EOB-MRI (P = 0.182) and CE-CT (P = 0.361), but the difference was not significant.

Non-SN type tumor diagnostic abilities and misclassification incidences for all imaging modalities
Sensitivities, specificities, accuracies, PPVs, and NPVs for diagnosis of non-SN type tumors are shown in   gross classification was an independent HCC prognostic factor in a nationwide study of Japanese patients [27], and Nakashima, et al., observed a higher incidence of intrahepatic metastasis and portal vein invasion in non-SN tumors compared to SN types [28]. Since non-SN type HCCs often invade within the tumor-bearing liver  segment, systematized or anatomic hepatectomy is generally performed [29]. Gross classification of HCCs before treatment could significantly improve patient outcomes.
In this study, we assessed the abilities of CE-CT, EOB-MRI, and CE-CT plus EOB-MRI to diagnose non-SN HCC. The AUC values for diagnosis of non-SN type HCCs by CE-CT and EOB-MRI separately were > 0.7, which corresponds to a moderate diagnostic value. However, the values for CE-CT combined with EOB-MRI were > 0.9, corresponding to a higher diagnostic value. This difference was significant compared to CE-CT alone, but not to EOB-MRI. EOB-MRI alone was a superior diagnostic method compared to CE-CT for all tumor sizes, although as tumor diameter increased, the ability of CE-CT to diagnose non-SN type tumors improved. In HCCs > 3 cm, AUC values were > 0.8 for CE-CT, EOB-MRI, and CE-CT plus EOB-MRI. These results suggest that CE-CT combined with EOB-MRI improves the accuracy of HCC gross classification. This may be because EOB-MRI offers high soft tissue contrast and superior spatial resolution, as well as hepatobiliary phase images. The EOB-MRI hepatobiliary phase may improve signal contrast between hypointense HCCs and the surrounding enhancing liver parenchyma, and thus tumor margins are clearly displayed [30].
In HCCs ≤ 3 cm, EOB-MRI diagnostic sensitivity and accuracy values in our study were similar to those reported by Toshifumi, et al. [ [31]. Simultaneous analyses of arterial, venous and equilibrium phases of thin-sliced CE-CT, as well as recent advances in imaging techniques, may contribute to the higher diagnostic accuracy observed in our study. Additionally, tumor specimens in our study were cut similar to the axial image section, and this may have helped improve CE-CT diagnostic accuracy.
Each imaging modality studied has limitations. For hypovascular HCCs, CE-CT cannot show tumor margins clearly. Since microvascular invasion or portal vein tumor thrombosis is present in some tumors, peritumoural liver parenchyma could reduce enhancement on hepatobiliary phase images, resulting in decreased tumor-liver contrast [32]. Furthermore, 10-20% of HCCs remain isointense or even hyperintense during the hepatocellular phase, which correlates with expression of organic anion-transporting polypeptides (OTAPs) in tumor cells [33,34]. It is challenging to clearly display tumor margins on EOB-MRI hepatobiliary phase images, and this can lead to misdiagnosis. In the current study, seven nodules were misclassified via CE-CT plus EOB-MRI, as a result of misdiagnosis of SN type tumors as non-SN, and these misclassified nodules were treated with anatomical hepatectomy. However, most nodules misclassified in this study resulted from misdiagnosis of non-SN type tumors as SN via CE-CT, especially in HCCs ≤ 3 cm. Similar results were reported by Tomoki, et al. [12]. This problem likely occurred because smaller tumor sizes result  in lower contrast and spatial resolution, and because of poor contrast-enhancement, particularly in CE-CT. In the present study, HCC gross classification via CE-CT plus EOB-MRI exhibited higher accuracy and sensitivity compared to either modality alone. The number of patients with MVI was greater in non-SN type (33.9%) than in SN type HCCs (13.2%, P = 0.01). Non-SN classification is an MVI predictor in HCCs and contributes to poor patient prognosis posthepatic resection or transplantation [7,22,28]. Anatomic hepatectomy or partial hepatic resection with a wide tumor margin is recommended to eradicate MVI [35,36]. Misclassification of non-SN type tumors as SN might negatively impact patient survival due to differences in treatments or surgical procedures resulting from misdiagnosis.
The present study has some limitations. First, ours was a single-center retrospective study with a relatively small patient population. Further studies with a larger number of patients are needed. Second, 92 of the study patients were Child-Pugh class A, which suggests they had good liver conditions. Enrolling more patients with poor liver conditions (Child-Pugh class B or C) might decrease the ability of EOB-MRI plus CE-CT to distinguish non-SN type tumors. Third, this study lacked prognosis information for a portion of the enrolled patients. It will essential to clarify tumor recurrence and patient survival rates for HCCs diagnosed as non-SN via EOB-MRI and/ or CE-CT in the future.
In conclusion, accurate gross classification through imaging is critical for determination of HCC patient prognoses and appropriate treatment strategies. HCC diagnosis via CE-CT plus EOB-MRI might provide a more accurate imaging assessment for HCC gross classification.

Patients
Study protocols were approved by our Institutional Review Board. Patient informed consent was waived due to the retrospective nature of the study. From July 2014 to May 2016, 153 consecutively registered patients with HCCs were subjected to curative hepatectomy at our institution. We based HCC diagnoses on postoperative pathology. Patients were included according to the following criteria: (1) gross specimen photographs could be reviewed; (2) no preoperative therapy was provided, such as radiofrequency or microwave ablation, transcatheter arterial chemoembolization (TACE), or molecular targeted therapy; (3) both CE-CT and EOB-MRI were preoperatively performed. Ninety-four patients meeting these criteria were enrolled in our study (Table 4). Another 59 patients were excluded for the following reasons: tumors with other cell types, such as cholangiocellular carcinoma (16 cases) or metastatic liver cancer (22 cases), and incomplete clinical data (21 cases). In patients with multiple HCCs, a major tumor was analyzed.
All patients were examined with multidetector spiral CT (LightSpeed VCT; GE Healthcare) in pre-contrast enhanced and multiphase contrast-enhancement scans with a slice thickness of 1.25 mm. Hepatic arterial phase, portal venous phase, and equilibrium phase images were obtained at 30, 60 and 180 s with 2 mL/kg bodyweight contrast media (Omnipaque 350 mgI/mL; GE Healthcare) injected through the antecubital vein (4 mL/s). Other abdominal CT scan parameters were as follows: tube voltage 120 kVp, tube current 240 mA, rotation time 0.6 s, helical pitch 1.375, field of view 35-40 cm, and matrix 512 × 512. HCC gross classification via CE-CT was based on the arterial, portal and equilibrium phases.

Pathological assessment of gross classification
A section of each resected tumor specimen was collected similarly to the section used for axial CT and MRI before immediate photographing and fixation in 10% formalin. In the current study, these photographs were used to retrospectively assess HCC gross classification. In light of the General Rules of the Clinical and Pathological Study of Primary Liver Cancer in Japan [2], two experienced pathologists and two surgeons divided pathological

Imaging assessment of gross classification
Through retrospective review of both CE-CT and EOB-MRI images, we categorized 94 nodules into five gross types, including SN-IM, SN-DM, SN-EG, (tumors were round with extranodular growth), CMN (tumors were lobulated with multiple nodular lesions), and IF (tumors were lobulated with an irregular margin). We then divided these five gross types into two groups: simple nodular (SN), including SN-IM and SN-DM, and non-simple nodular (non-SN), including SN-EG, CMN, and IF, because the latter had higher malignant potential than the former. Preoperative CE-CT and EOB-MRI images were reviewed by one hepatologist (Xu Fu, with four years of experience) and one radiologist specializing in hepatology (Min Tang, with 10 years of experience) independently; both evaluators were blinded to patient data. They independently evaluated hepatobiliary phase EOB-MRI axial images, as well as CE-CT arterial, venous and equilibrium phase axial images to predict the gross classification of nodules. The readers independently made their classifications and in cases of disagreement a consensus was reached after discussion.

Statistical analysis
Data were presented as means ± standard deviation or median (range). The Mann-Whitney U test was performed to compare continuous variables, and Fisher's exact test compared categorical variables. The receiver operating characteristic (ROC) curve was applied to evaluate the diagnostic performance of CE-CT, EOB-MRI and both combined [38]. An area under the ROC curve (AUC) of > 0.90 corresponded to high accuracy, 0.70-0.90 indicated moderate accuracy, and 0.50-0.70 low accuracy [39]. The Chi squared test compared sensitivities, specificities, accuracies, positive predictive values (PPV), and negative predictive values (NPV). For the evaluation of interobserver variability with respect to classified imaging findings, kappa statistics was conducted. Agreement was considered to be almost perfect at a kappa value of 1.00-0.81; good at 0.80-0.61; moderate at 0.60-0.41; fair at 0.40-0.21; and poor at < 0.21 [40]. All data were analyzed using SPSS19.0 statistical software (SPSS, Chicago, IL, USA) and MedCalc11.4.2 software (Ostend, Belgium). P < 0.05 indicated statistical significance.