Introduction

Gadoxetic acid–enhanced magnetic resonance imaging (GA-MRI) is one of the primary imaging modalities for the evaluation of patients with hepatocellular carcinoma (HCC) [1,2,3]. GA-MRI has been reported to have a higher sensitivity than extracellular contrast-enhanced MRI in the detection of HCC [4], probably because GA-MRI can detect additional nodules on hepatobiliary phase (HBP) images [5,6,7]. HBP images are obtained approximately 20 min after the injection of gadoxetic acid when maximal hepatic parenchymal enhancement is reached, according to the uptake of contrast media by normal, functioning hepatocytes [8]. Because pathologic lesions such as HCCs lack functional hepatocytes and do not uptake the contrast media and appear hypointense compared to surrounding hepatic parenchyma [9], HBP images provide excellent lesion-to-liver contrast that aids in the detection of such lesions. In addition to its diagnostic role, HBP images have been recently reported to have prognostic value for patients with HCC [10,11,12]. For instance, non-smooth tumor margins or peritumoral hypointensity on HBP images has been correlated with pathological microvascular invasion (MVI), which is associated with early recurrence after surgical resection or radiofrequency ablation (RFA) in patients with HCC [10, 13].

It can be hypothesized that there might be prognostic factors specific to patients with HCC receiving different treatments, and those factors may be captured in the HBP of GA-MRI. Obtaining prognostic information from as well as evaluating the presence or absence of HCC on HBP images may be clinically useful for choosing a treatment method and predicting the prognosis of patients with HCC [14]. For example, if there is an HBP feature that is specifically associated with poor prognosis after RFA, detecting that feature may lead physicians to consider alternatives such as resection or transplantation in that patient. The clinical impact of the prognostic value of HBP features might be enhanced by the increasing popularity of abbreviated GA-MRI protocols that rely on HBP images for the surveillance of HCC [15]. However, to the best of our knowledge, the prognostic value of HBP features for patients with HCC receiving different treatment methods has not been investigated.

Therefore, we aimed to identify the clinical findings and HBP features of GA-MRI with prognostic value in patients with HCC who underwent different treatments, including resection and locoregional therapy.

Materials and methods

This retrospective study was approved by the Institutional Review Board of Seoul National University Hospital (IRB No. 1901-036-1001), and the requirement for informed consent was waived.

Patients

We retrospectively included consecutive patients with HCC who underwent GA-MRI at Seoul National University Hospital between January 2014 and December 2015. The inclusion criteria were as follows: (a) patients with HCC who met the non-invasive diagnostic criteria (i.e., arterial phase hyperenhancement and washout in the portal venous phase) [1, 2]; (b) patients with very early/early stage HCC according to the Barcelona Clinic Liver Cancer (BCLC) staging system [2]; and (c) patients with an interval between GA-MRI and treatment < 1 month. The exclusion criteria were as follows: (a) patients with HCC beyond the early stage; (b) patients who underwent treatment for HCC before the GA-MRI examination; (c) patients with a pathologic diagnosis after surgical resection that was not HCC; (d) patients who had other malignancies; and (e) patients who underwent liver transplantation or percutaneous ethanol injection after GA-MRI. We excluded patients who underwent transplantation or percutaneous ethanol injection because the numbers of those patients were relatively small (n = 10 and 9, respectively). Initially, 347 patients underwent the following treatments: resection (n = 130), RFA (n = 96), and transarterial chemoembolization (TACE) (n = 121) [16, 17]. Then, propensity score analysis using logistic regression and 1:1 matching among treatment groups was performed to reduce the effect of potential selection biases and confounding factors [18]. The variables used for propensity score matching were age, sex, etiology of liver disease, Child-Pugh score, serum alpha-fetoprotein (AFP) level, and tumor size and number on GA-MRI according to the radiologic reports. The final cohort comprised 183 patients with 183 HCCs (n = 61 for each treatment) (Fig. 1). Laboratory findings including the serum levels of albumin, total bilirubin, and AFP and prothrombin time-international normalized ratio (PT-INR) were gathered from the electronic medical records. For the patients who underwent resection, the presence or absence of MVI in the surgical specimens was recorded.

Fig. 1
figure 1

Flow diagram of the study population. HCC hepatocellular carcinoma, GA-MRI gadoxetic acid–enhanced magnetic resonance imaging, RFA radiofrequency ablation, TACE transarterial chemoembolization

Full size image

GA-MRI examination

GA-MRI was performed using either a 3.0-T or 1.5-T system. The liver GA-MRI protocol used in our institution included a breath-hold fat-saturated T2-weighted fast spin-echo or turbo spin-echo sequence, a three-dimensional T1-weighted dual-echo sequence, diffusion-weighted imaging, and dynamic three-dimensional fat-saturated T1-weighted gradient-echo sequences including HBP imaging. The details of our GA-MRI examinations are described in the Supplementary Material.

HBP image analysis

Two radiologists (D.H.L. and J.H.K., with 14 and 7 years of clinical experience in reading liver MRIs, respectively) who were informed that the patients had very early/early stage HCC but were blinded to the other clinical and pathological information independently evaluated the imaging features of HCC and liver function on HBP images. Although image analysis was performed only on HBP images, the radiologists could refer to any other pulse sequences to locate the HCC on the HBP images. The following HBP features of HCC were evaluated: number, size, signal intensity, heterogeneity, margin, capsule, peritumoral hypointensity, and satellite nodules. Tumor margins were graded as smooth or non-smooth [19]. The capsules were graded as absent, hypointense, or hyperintense [20]. Peritumoral hypointensity was defined as an irregular, wedge-shaped or flame-like hypointense area of liver parenchyma located outside of the tumor margin and was classified as absent or present [21]. Satellite nodules were defined as nodules < 1 cm in diameter outside the tumor margins within the venous drainage region, within 2 cm from the tumor, and were classified as absent or present [22]. To assess liver function, contrast between the liver and portal vein and biliary excretion into the common bile duct were evaluated [23, 24]. When there were multiple tumors, the largest tumor was assessed. Further details of the imaging feature analyses are described in the Supplementary Material. After independently reviewing the images, the two radiologists reached a consensus, and discrepancies were resolved through a discussion with another senior radiologist (J.H.K., with 22 years of clinical experience in reading liver MRIs).

Treatment and follow-up

The treatment method was chosen according to the European Association for the Study of the Liver guidelines for patients with very early/early stage HCC and included resection, RFA, or transplantation [2]. In our institution, however, TACE is frequently performed as a first-line treatment for patients who cannot undergo RFA due to poor tumor visibility on ultrasound or a risky tumor location (i.e., abutting the hilar bile duct) [16, 17]. Because we excluded patients who underwent transplantation, our patients underwent one of the following three treatment methods: resection, RFA, and TACE. For each patient, the treatment method was chosen at the clinician’s discretion considering age, general condition, and preference of the patient. Follow-up data after treatment were obtained by reviewing the electronic medical records of our institution and by accessing statistical data from the Korean Ministry of Government Administration and Home Affairs. Disease-free survival (DFS) and overall survival (OS) were defined as the interval between the date of treatment and the date of any kind of tumor progression or the date of death or last follow-up visit before July 31, 2019, respectively. The details of the follow-up protocol and assessments of recurrence are described in the Supplementary Material.

Statistical analysis

The baseline characteristics and HBP features of the patients were compared by using the chi-square test or Fisher’s exact test for categorical variables and independent t test or the Mann-Whitney U test for continuous variables. Bonferroni correction was used for multiple pairwise comparisons. In the resection group, the associations between MVI and HBP features were assessed by using the chi-square test or Fisher’s exact test. Linearly weighted kappa statistics and intraclass correlation coefficients were used to assess interobserver agreement for categorical and continuous variables, respectively. To interpret the interobserver agreement results, the following convention was used: < 0.20, poor; 0.21–0.40, fair; 0.41–0.60, moderate; 0.61–0.80, substantial; and 0.81–1.00, nearly perfect [25]. Survival curves were drawn by using the Kaplan-Meier method. Patients who underwent liver transplantation after the initial treatment were censored at the date of transplantation. The cumulative incidence of tumor recurrence was also calculated by using the Kaplan-Meier method. A univariable Cox proportional hazard model was used to reveal the clinical or HBP imaging features associated with OS and DFS. The variables with p < 0.050 were included in the multivariable Cox proportional hazard model. A p value < 0.050 indicated statistical significance. All statistical analyses were performed using commercially available software (IBM SPSS Statistics for Windows, version 25, IBM; MedCalc, version 18.2.1).

Results

Baseline characteristics of the patients

The baseline characteristics of the patients before and after propensity score matching are summarized in Table S2 and Table 1, respectively. Among these patients, 75 patients had very early stage HCC, and the remaining 108 patients had early stage HCC. The interval between GA-MRI and treatment was 11.8 ± 9.1 days (range, 0–31 days). In the resection group, MVI was observed in 18.0% (11 of 61 patients) of the surgical specimens. There were significant differences in DFS and OS among the three treatment groups.

Table 1 Clinical characteristics of study population
Full size table

HBP features according to different treatments

The HBP features are summarized in Table 2. The contrast between the liver and portal vein was relatively preserved in the resection group compared with that in the RFA or TACE group (isointense:slightly hyperintense:hyperintense, 1:9:51 vs. 1:22:38 vs. 2:23:36, p = 0.033). Otherwise, there was no significant difference in the distribution of HBP features among the treatment groups. In terms of MVI in the resection group, the tumors with MVI were significantly larger than those without MVI (median, 2.8 cm vs. 2.4 cm, p = 0.042). The other HBP features did not demonstrate differences regarding the presence of MVI (p > 0.050). Interobserver agreement for the HBP features ranged from poor to nearly perfect (0.173–0.951).

Table 2 Distribution of HBP features according to the treatment
Full size table

Clinical and HBP features with prognostic value for DFS in patients receiving different treatments

After receiving their treatments, the patients were followed up for a median (range) and mean ± standard deviation (SD) period of 51 months (5–66 months) and 47.3 ± 14.3 months, respectively. The median (range) and mean ± SD DFS were 32 months (2–65 months) and 31.0 ± 19.9 months, respectively. During surveillance after treatment, 14 patients were lost to follow-up. The estimated 1-, 3-, and 5-year DFS rates after treatment for HCC were 74.9%, 51.8%, and 40.0%, respectively. The patients in the resection group (47.7 months) and RFA group (39.0 months) had a better mean DFS than those in the TACE group (26.2 months) (p < 0.001 and p = 0.001, respectively), whereas there was no significant difference in mean DFS between the resection group and RFA group (p = 0.124). Of the 95 patients in whom recurrence was detected, local tumor progression developed in 25, intrahepatic distant recurrence developed in 68, and extrahepatic metastasis developed in 2. When assessed according to treatment group, recurrence (local tumor progression/intrahepatic distant recurrence/extrahepatic metastasis) occurred in 22 (1/20/1), 30 (4/26/0), and 43 (20/22/1) patients in the resection, RFA, and TACE groups, respectively. In terms of the results of the multivariable Cox proportional hazard model, in the resection group, tumor size (hazard ratio [HR] 4.159 per centimeter; 95% confidence interval [CI], 1.669–10.365) on HBP images was the only feature associated with DFS (Table S3). In the RFA group, serum albumin level (HR 0.390; 95% CI, 0.166–0.917) and satellite nodules on HBP images (HR 5.037; 95% CI, 1.061–23.903) were associated with DFS (Fig. 2) (Table S4). Finally, in the TACE group, no clinical factors or HBP features were significantly associated with DFS (Table S5).

Fig. 2
figure 2

Representative images of patients without or with satellite nodules on hepatobiliary phase images. a This 61-year-old man had a 44-mm-sized hepatocellular carcinoma without satellite nodules. This patient did not experience recurrence for 50 months after radiofrequency ablation and was censored. b This 73-year-old man had a 42-mm-sized hepatocellular carcinoma but with a satellite nodule (arrow). Radiofrequency ablation was performed for this patient; however, intrahepatic distant recurrence was detected 12 months after treatment. c Kaplan-Meier curves demonstrated a significant difference in disease-free survival between patients without satellite nodules and those with satellite nodules (p = 0.002)

Full size image

Clinical and HBP features with prognostic value for OS in patients receiving different treatments

The estimated 1-, 3-, and 5-year OS rates after treatment for HCC were 98.4%, 85.8%, and 82.0%, respectively. During the follow-up period, 30 of the 183 patients (16.4%) died because of HCC progression (n = 14), liver failure (n = 6), variceal bleeding (n = 2), septic shock (n = 2), or an unknown cause (n = 6). Three of the 183 patients (1.6%) underwent liver transplantation for HCC recurrence after a median follow-up of 27 months (range, 13–58 months). There was no significant difference in mean OS among the treatment groups (p = 0.062).

In the resection group, a large tumor size on HBP images was associated with poor OS (HR 8.495 per centimeter; 95% CI, 1.072–67.338) (Table 3). In the RFA group, peritumoral hypointensity (HR 13.062; 95% CI, 1.627–104.840) and satellite nodules (HR 9.398; 95% CI, 1.480–59.668) on HBP images were associated with poor OS (Fig. 3) (Table 4). Serum albumin level (HR 0.064; 95% CI, 0.006–0.687) and PT-INR (HR 1413.55; 95% CI, 2.245–889885) were also associated with OS. In the TACE group, no clinical factors or HBP features were significantly associated with OS (Table 5).

Table 3 Clinical and HBP features of prognostic value for overall survival in the resection group
Full size table
Fig. 3
figure 3

Representative images of patients without or with peritumoral hypointensity on hepatobiliary phase images. a This 74-year-old man had a 20-mm-sized hepatocellular carcinoma without peritumoral intensity. This patient survived 46 months after radiofrequency ablation and was censored. b This 40-year-old woman also had an 18-mm hepatocellular carcinoma with peritumoral intensity (arrow). This patient died 25 months after radiofrequency ablation. c Kaplan-Meier curve demonstrated a significant difference in overall survival between patients without peritumoral intensity and those with peritumoral intensity (p = 0.013)

Full size image
Table 4 Clinical and HBP features of prognostic value for overall survival in the RFA group
Full size table
Table 5 Clinical and HBP features of prognostic value for overall survival in the TACE group
Full size table

Discussion

This study demonstrated that there were different clinical factors and HBP features of prognostic value specific to patients with HCC receiving different treatments. In the resection group, a large tumor size was associated with poor DFS/OS. In the RFA group, satellite nodules and/or peritumoral hypointensity on HBP images, decreased serum albumin levels, and/or elevated PT-INR were prognostic factors for poor DFS/OS. Finally, in the TACE group, no clinical factors or HBP features were significantly associated with DFS/OS. These results may serve as a reference when making treatment decisions. For instance, either RFA or resection would be recommended for a very early stage single HCC < 2 cm according to the BCLC staging system. However, our results indicate that resection might be preferred over RFA in that patient to improve OS if peritumoral hypointensity is observed on the HBP images. The prognostic implication of HBP features may be associated with MVI, which is a well-known risk factor for early recurrence and/or poor OS in patients with HCC [26]. All the HBP features that demonstrated prognostic value for DFS/OS in our study (i.e., large tumor size, satellite nodules, and peritumoral hypointensity) have been reported to be associated with MVI [27]. Interestingly, different HBP features of MVI showed different associations with different treatment methods. For example, peritumoral hypointensity was associated with OS in the RFA group only but not in the resection group or TACE group, whereas large tumor size showed prognostic value in the resection group only. This finding raises the possibility that different MVI-associated HBP features might have different mechanisms by which they appear on HBP images, which could have different therapeutic implications. Although there was no significant association between HBP features and MVI in this study, we speculate that the relatively small number of patients in the resection group (n = 61) who were pathologically confirmed to have MVI may be responsible for the statistical insignificance. Further studies are warranted to elucidate the mechanism and relationship between HBP features and MVI.

Other HBP features, including a non-smooth tumor margin, tumor encapsulation, and high/isointense signal in the tumor, did not demonstrate an association with DFS/OS in our study. Although a few studies advocated for the association of these HPB features with MVI or their prognostic role in patients with HCC [12, 28,29,30], controversy remains in the literature [31, 32]. Regarding tumor encapsulation, Kim et al reported that a hypointense rim around the tumor on HBP may not necessarily indicate a tumor capsule or pseudocapsule [32, 33]. In terms of tumor signal intensity, only a small proportion of our tumors showed iso- or hyperintensity on HBP images (4.4% [8/183]), and this small number may have resulted in the lack of association between tumor signal intensity and prognosis.

There have been studies on the prognostic values of other pulse sequences of gadoxetic acid–enhanced MRI, such as enhancement patterns on dynamic T1-weighted images or diffusion-weighed imaging. In patients who underwent surgical resection for HCC, a low apparent diffusion coefficient map value has been reported to be associated with poor prognosis after resection [34, 35]. According to a recent study by Rhee et al, the arterial phase enhancement pattern is associated with macrotrabecular-massive HCC, which is a subtype of HCC that has a poor prognosis [36]. Other non-HBP imaging features, such as peripheral ragged enhancement, have been demonstrated to have prognostic value in patients who underwent TACE [37, 38].

In addition to the HBP features, a few clinical findings were also associated with DFS/OS in the RFA group: serum albumin level and PT-INR. Because both factors are associated with liver function [39, 40], the association of these laboratory results with DFS/OS in patients with HCC may reflect the importance of liver function. In the resection group and TACE group, however, no clinical findings or HBP features of liver function were associated with DFS/OS. One possible explanation for this result may be the inclusion criteria of this study: patients with preserved liver function (i.e., BCLC very early/early stage) were assessed. Moreover, propensity score matching for tumor number and size led to the selection of patients with single tumors and small HCCs. This selection criterion had a greater impact on the resection and TACE groups than on the RFA group because the resection group had larger tumors and the TACE group had more tumors than the RFA group before propensity score matching.

There were several limitations in this study. First, selection bias may have been present in our study because it is a single-center, retrospective study. In addition, there is a possibility that these HBP features might have affected treatment decision-making, although there was no direct evidence that these HBP features were used to select the treatment method. However, we minimized this selection bias by including only treatment-naïve patients with very early/early stage HCC. Second, only HBP images were used, and information from dynamic studies, T2-weighted imaging, or diffusion-weighted imaging was not available. This situation may be problematic when diagnosing HCC. However, we focused on the prognostic role of HBP images, not on the diagnosis of HCC. Finally, an assessment of the association between MVI and HBP features was only possible in the resection group because the RFA and TACE groups lacked pathological information.

In conclusion, different HBP features and clinical factors were specifically associated with DFS/OS in patients with HCC receiving different treatment methods. The identified HBP features may have the potential to predict prognosis and select treatment methods for patients with HCC.