Note: This copy is for your personal non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at www.rsna.org/rsnarights.

Original Research  n  Gastrointestinal

Single Hepatocellular Carcinoma: Preoperative MR Imaging to Predict Early Recurrence after Curative Resection1 Purpose:

To identify magnetic resonance (MR) imaging features that enable prediction of early recurrence (,2 years) after curative resection of hepatocellular carcinoma (HCC) and to derive a preoperative prediction model.

Materials and Methods:

This retrospective study was approved by the institutional review board. The requirement to obtain written informed consent was waived. A total of 268 patients who underwent hepatic resection for a single HCC from January 2008 to August 2011 were divided into two cohorts: a training cohort, which was used to derive a prediction model (n = 187), and a validation cohort (n = 81). All MR images from the training cohort were reviewed by two radiologists. A prediction model was constructed by using MR imaging features that were independently associated with early recurrence with use of multiple logistic regression analysis. The performance of the prediction model in the validation cohort was evaluated with respect to discrimination (ie, whether the relative ranking of individual predictions of subsequent early recurrence is in the correct order).

Results:

In the training cohort, four MR imaging features were independently associated with early recurrence: rim enhancement (odds ratio [OR] = 3.83; 95% confidence interval [CI]: 1.39, 10.52), peritumoral parenchymal enhancement in the arterial phase (OR = 2.64; 95% CI: 1.27, 5.46), satellite nodule (OR = 4.07; 95% CI: 1.09, 15.21), and tumor size (OR = 1.66; 95% CI: 1.31, 2.09). A prediction model derived from these variables showed an area under the receiver operating characteristic curve (AUC) of 0.788 in the prediction of the risk of early recurrence in the training cohort. When applied to the validation cohort, this model showed good discrimination (AUC, 0.783).

Conclusion:

The prediction model derived from rim enhancement, peritumoral parenchymal enhancement, satellite nodule, and tumor size can be used preoperatively to estimate the risk of early recurrence after resection of a single HCC.

1

 From the Department of Radiology, Research Institute of Radiological Science (C.A., Y.E.C., H.R., M.J.K.), and Department of Pathology (Y.N.P.), Severance Hospital, Yonsei University College of Medicine, 50 Yonsei-Ro, SeodaemunGu, Seoul 120-752, South Korea; and Department of Policy Research Affairs, National Health Insurance Corporation Ilsan Hospital, Goyang, Korea (D.W.K.). Received October 10, 2014; revision requested November 17; revision received December 8; accepted December 16; final version accepted December 18. Address correspondence to M.J.K. (e-mail: [email protected]).

Imaging

Chansik An, MD Dong Wook Kim, PhD Young-Nyun Park, MD Yong Eun Chung, MD Hyungjin Rhee, MD Myeong-Jin Kim, MD

 RSNA, 2015

q

Online supplemental material is available for this article.

 RSNA, 2015

q

Radiology: Volume 000: Number 0—   2015  n  radiology.rsna.org

1

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

H

epatic resection is the primary treatment modality for hepatocellular carcinoma (HCC) in patients with well-preserved liver function (1,2). However, recurrence rates after hepatic resection can be as high as 50% within 5 years (3–5). More than 80% of recurrent tumors develop in the remnant liver, and the mechanisms of this intrahepatic recurrence can be either intrahepatic metastasis from the initial tumor or de novo multicentric occurrence (6). Of these, intrahepatic metastasis is associated with worse prognosis and usually manifests as early recurrence within the first 2 years after resection of HCC (7–11). Therefore, the risk factors for early recurrence after hepatic resection have been extensively studied. Early recurrence is more likely to be associated with tumor factors such as microvascular invasion, worse histologic differentiation, and microsatellite nodules, whereas late recurrence is more likely related to underlying liver conditions such as the presence of cirrhosis (9,10,12–15). However, most of these tumor factors can only be evaluated with postoperative pathologic examination and thus may not be useful for deciding treatment before hepatic resection. The close relationship between early recurrence and tumor factors implies

Advances in Knowledge nn Larger tumor size (odds ratio [OR] = 1.66) and the presence of satellite nodules (OR = 4.07), rim arterial enhancement (OR = 4.07), and peritumoral parenchymal enhancement (OR = 2.64) at preoperative MR imaging are independent predictors of early recurrence (,2 years) after curative resection of a single hepatocellular carcinoma (HCC). nn A prediction model derived from these MR imaging features could be used to estimate the likelihood of early recurrence after curative resection of a single HCC (area under the receiver operating characteristic curve, 0.783). 2

that preoperative imaging findings of HCC could be used to predict early recurrence after hepatic resection. Findings such as peritumoral parenchymal enhancement at magnetic resonance (MR) imaging or computed tomography (CT) and peritumoral hypointensity and/or irregular tumor margin in the hepatobiliary phase of MR imaging may be useful for predicting microvascular invasion, a known risk factor for early recurrence (16–21). Other imaging features, such as rim enhancement in the arterial phase of dynamic CT or MR imaging (22,23) and an irregular tumor margin and signal intensity on hepatobiliary images at gadoxetic acid–enhanced MR imaging, have been suggested as useful findings to predict tumor behavior (7,18,24). To our knowledge, however, the utility of the combination of those findings for predicting postoperative recurrence has not been studied. The purpose of this study was to identify MR imaging features that enable the prediction of early recurrence (,2 years) after curative resection of HCC and to derive a preoperative prediction model.

Materials and Methods

An et al

included. Specifically, HCC was considered single when nodules close to the main tumor were described as satellite nodules in the original pathology report. HCCs were considered to be multiple if two or more HCCs were reported separately with a full description of histopathologic features such as histologic grade, architecture, and cell type. Patients were excluded from the study if they had received local-regional therapy such as chemoembolization or chemoradiation before surgery (n = 105), were followed up for less than 2 years (n = 25), died of postoperative complications within 2 weeks (n = 7), or had ruptured HCC (n = 2). Therefore, the final study population included 268 patients with 268 HCCs. The median time between MR imaging and surgery was 12.5 days (range, 1–60 days). For analysis, the first 187 patients who underwent hepatic resection from January 2008 to September 2010 were allocated into the training cohort to derive a prediction model; the subsequent 81 patients were allocated into the validation cohort. Clinical information, laboratory data, and pathology reports were reviewed retrospectively from electronic

Published online before print 10.1148/radiol.15142394  Content code:

Patients This retrospective study was approved by our institutional review board, and the requirement to obtain written informed consent was waived. From January 2008 to August 2011, 546 consecutive patients underwent curative hepatic resection for HCC at our institution. Of these, 407 patients who underwent gadoxetic acid–enhanced MR imaging within 2 months before surgery and in whom postoperative pathologic examination showed a single HCC regardless of the presence of satellite nodules were Implication for Patient Care nn The ability to preoperatively estimate the risk of early recurrence after curative resection may be clinically relevant for guiding the treatment of patients with HCC.

Radiology 2015; 000:1–11 Abbreviations: ADC = apparent diffusion coefficient AFP = a-fetoprotein AUC = area under the receiver operating characteristic curve CI = confidence interval HCC = hepatocellular carcinoma OR = odds ratio PIVKA-II = protein induced by vitamin K absence or antagonist-II Author contributions: Guarantors of integrity of entire study, C.A., D.W.K., M.J.K.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; agrees to ensure any questions related to the work are appropriately resolved, all authors; literature research, C.A., Y.N.P., M.J.K.; clinical studies, C.A., Y.N.P., Y.E.C., H.R., M.J.K.; statistical analysis, C.A., D.W.K.; and manuscript editing, C.A., Y.N.P., Y.E.C., M.J.K. Conflicts of interest are listed at the end of this article.

radiology.rsna.org  n Radiology: Volume 000: Number 0—   2015

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

An et al

Table 1 MR Imaging Parameters Sequence* Dual-echo T1-weighted GRE

T1-weighted 3D GRE

T2-weighted TSE, navigator-triggered

Diffusion-weighted imaging

Field Strength (T)†

Matrix

1.5 (Intera Achieva) 3.0 (Magnetom Trio) 3.0 (Intera Achieva) 1.5 (Intera Achieva) 3.0 (Magnetom Trio) 3.0 (Intera Achieva) 1.5 (Intera Achieva) 3.0 (Magnetom Trio) 3.0 (Intera Achieva)

256 3 256 256 3 192 222 3 172 256 3 256 256 3 192 268 3 266 288 3 230 256 3 192 400 3 328 or  352 3 329 128 3 100 192 3 108 124 3 124

1.5 (Intera Achieva) 3.0 (Magnetom Trio) 3.0 (Intera Achieva)

Section Thickness (mm)

Intersection Gap (mm)

Repetition Time (msec)

Echo Time (msec)

Flip Angle (degrees)

7 6 7 4 2–2.5 3 7 5 or 4 4 or 7

7.7 6.6–7.2 7.7 2 2–2.5 1.5 5 6 or 5 5 or 8

167 150–171 192 4.48 2.54 2.99 2000 or 452 3250–4770 or 466 1636 or 1589

2.3 and 4.6 2.46 and 1.23 2.3 and 1.14 2.2 0.92–0.95 1.41 80 88–96 70 or 80

80 65 50 15 11.5–13 10 90 140–150 90

7 5 5

8 6 5.5

2139–2691 1500–6100 1634

44–53 69–70 56

90 90 90

* GRE = gradient-recalled echo, 3D = three-dimensional, TSE = turbo spin echo. †

Trade name is in parentheses.

medical records. Clinical information included patient demographics, cause of chronic liver disease, and time to recurrence. Early recurrence was defined as recurrence within 2 years after curative resection of HCC. Laboratory data included a-fetoprotein (AFP) and protein induced by vitamin K absence or antagonist-II (PIVKA-II) levels. At our institution, the reference values of AFP and PIVKA-II are less than 9 ng/mL and less than 40 mAU/mL, respectively. Pathologic data analyzed in this study were tumor size, presence of satellite nodules defined as microscopic nodules of HCC separated from the tumor by an interval of uninvolved liver parenchyma, histologic differentiation, surgical resection margin (invasion or free of tumor), presence of serosal invasion, and presence of gross and/or microscopic vascular invasion. All of this information was taken from the original surgical pathology reports that were issued by an experienced pathologist (Y.N.P., with .20 years of experience in liver histopathology) who reviewed and completed preliminary reports by pathology residents.

MR Imaging MR imaging was performed by using a 3.0-T system (Magnetom Trio a Tim [Siemens Medical Solutions, Erlangen,

Germany] or Intera Achieva [Philips Medical Systems, the Netherlands]) or a 1.5-T system (Intera Achieva). All MR imaging parameters are listed in Table 1. All images were obtained in the transverse plane with a field of view of 44 3 33 cm or 40 3 30 cm according to the patient’s body size. After localizer images were obtained, two-dimensional dual-echo T1-weighted gradient-recalled-echo images were obtained (in phase and opposed phase). Dynamic images were obtained before and after contrast material administration in arterial, portal venous, hepatic venous, and final dynamic phases by using a threedimensional gradient-echo sequence. To determine the imaging delay for arterial phase imaging, a bolus technique was used with 1 mL of gadoxetic acid disodium (Primovist; Bayer Schering Pharma, Berlin, Germany) and a 20-mL 0.9% saline chaser at an injection rate of 1 or 2 mL/sec to determine the peak enhancement of the abdominal aorta. For dynamic imaging, 0.1 mL/kg (0.025 mmol/kg) of gadoxetic acid disodium was injected, followed by a 20-mL saline chaser at the same rate as that used for the bolus injection. The arterial phase began 2 or 3 seconds after the peak aortic enhancement was determined with the bolus injection; subsequent

Radiology: Volume 000: Number 0—   2015  n  radiology.rsna.org

dynamic images were obtained at intervals of approximately 30 seconds. Each dynamic image acquisition required 18–24 seconds. Hepatobiliary images were obtained 15 or 20 minutes after injection of the contrast material by using the same imaging sequence as that used for the pre- and postcontrast images. During the interval between dynamic and hepatobiliary phase imaging, T2-weighted images were obtained with multishot and single-shot turbo spin-echo sequences by using a navigator-triggered technique. Diffusionweighted images were also obtained by using a navigator-triggered technique at b values of 50, 400, and 800 sec/mm2, and the apparent diffusion coefficient (ADC) was automatically calculated by the MR units and displayed as a corresponding ADC map.

Image Analysis for the Training Set MR images were retrospectively reviewed with a local picture archiving and communication system (Centricity; GE Healthcare, Milwaukee, Wis). One radiologist (C.A.) first recorded the location (liver segment) and corresponding image number of each HCC by reviewing the pathology records, surgical records, and preoperative MR images. With this information, two other 3

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

Figure 1

Figure 1:  Rim arterial enhancement. Pathologically confirmed HCC in 61-year-old man. (a) Unenhanced and (b) gadoxetic acid–enhanced T1-weighted MR images in arterial phase show 4.2-cm tumor with irregular peripheral arterial enhancement and relatively hypovascular central portions.

Figure 2

Figure 2:  Peritumoral parenchymal enhancement. Pathologically confirmed HCC in 50-year-old man. Gadoxetic acid–enhanced T1-weighted MR images obtained in (a) arterial and (b) venous phases show this 3.5-cm tumor has typical enhancement pattern of arterial enhancement and venous washout as well as peritumoral wedge-shaped parenchymal arterial enhancement (arrow), which disappears in venous phase.

radiologists (Y.E.C. and H.R., with 11 and 4 years of experience in abdominal MR imaging, respectively), who were blinded to clinical, laboratory, and pathologic information, qualitatively and independently determined whether the following eight MR imaging features were present in each HCC from the training set: (a) peripheral rim enhancement in the arterial phase (Fig 1), defined as the presence of irregular ringlike areas of enhancement with 4

central hypovascular areas in the arterial phase, as described previously (25); (b) peritumoral parenchymal enhancement in the arterial phase (Fig 2), defined as grossly hyperarterial contrast material enhancement outside of the tumor border that becomes isointense with background liver parenchyma in the later dynamic phase images, regardless of shape (eg, wedge shaped or circumferential) (this definition was modified from one used in a previous

An et al

study [17]); (c) capsule, defined as a distinct low-signal-intensity ring with delayed contrast enhancement along the tumor border that involves more than 90% of the tumor circumference; (d) peritumoral low signal intensity in the hepatobiliary phase, defined as an irregular, wedge-shaped, or flamelike area of low signal intensity in the liver parenchyma located outside of the tumor margin in the hepatobiliary phase (21); (e) irregular tumor margin in the hepatobiliary phase, which is considered to be present when a tumor had a budding portion at its periphery protruding into the liver parenchyma (18); (f) satellite nodules, which are lesions smaller than 2 cm with similar MR imaging characteristics and located within 2 cm of the main tumor (26); (g) intratumoral fat, defined as an area of tumor with a decrease in signal intensity on opposed-phase T1-weighted images compared with in-phase images; and (h) gross vascular invasion, defined as invasion of the adjacent hepatic arteries, hepatic veins, or portal veins grossly visible on images. After the first independent image analysis, interobserver agreement for the assessment of the MR imaging features was evaluated. The two reviewers then met to draw final conclusions by consensus on discordant results. Quantitative image analysis was performed by a radiologist (C.A.) who measured the following four parameters for each HCC in the plane in which the tumor had its largest cross-sectional diameter: (a) tumor size, defined as the maximum diameter of each tumor as measured with the electronic caliper on the picture archiving and communication system. (b) Tumor-to-liver signal intensity ratio, which was calculated by dividing the signal intensity of the tumor by that of the liver. Regions of interest were placed as large as possible within the confines of the tumor for measuring the mean signal intensity of HCCs. To minimize measurement errors, we used an open-source imaging processing package (ImageJ, www.rsb.info. nih.gov) to copy and paste a region of interest drawn on hepatobiliary phase

radiology.rsna.org  n Radiology: Volume 000: Number 0—   2015

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

images onto the coregistered unenhanced T1-weighted images. The signal intensity of adjacent liver parenchyma was measured by using a round region of interest (100 mm2 in area) while excluding artifacts and blood vessels. (c) Signal intensity ratio on diffusionweighted images (b = 800 sec/mm2), where the tumor-to-liver signal intensity ratio on high-b-value diffusionweighted images was calculated with the same method as described earlier. (d) ADC values, which were obtained by manually drawing a region of interest on the ADC map to encompass the entire tumor, referencing T2-weighted transverse images.

Image Analysis for the Validation Set To validate the prediction model established from the results of the first review, the same two radiologists analyzed MR images from patients in the validation cohort by using the same method described earlier. For this second review session, they only assessed MR imaging features that had been independently associated with early recurrence in the training cohort. Quantitative image analysis was not performed in the validation cohort because none of the quantitative items was significantly associated with early recurrence in the training cohort. Statistical Analysis To compare variables between the patients in the training and validation sets, we used the Student t or MannWhitney U test for continuous variables and the x2 or Fisher exact test for categoric variables. The x2 or Fisher exact test was used to examine the relationships between pathologic and MR imaging findings. Univariate and multivariate logistic regression analyses were used to identify independent determinants of early HCC recurrence and microvascular invasion. For multivariate regression, multicollinearity was assessed by calculating variation inflation factors and condition indexes to determine whether independent variables showed significant intercorrelation. A prediction model was derived on the

An et al

Figure 3

Figure 3:  Nomogram to predict probability of recurrence within 2 years after curative resection of single HCC. Top, predictor points are found on uppermost point scale that corresponds to each variable. Bottom, points of all variables are added up, and total point projected on bottom scale indicates probability of early recurrence.

basis of multivariate logistic regression analysis. A nomogram was constructed on the basis of this prediction model (Fig 3). The estimated effects of the independent variables were ranked according to the estimated b coefficients from the logistic model. After we determined which variable had the greatest effect in the model and assigned 100 points to it, other variables were then sequentially assigned points on the basis of their proportions to the point assigned to the variable with the greatest effect. The nomogram is used by first locating the position of an HCC on each predictor variable scale. By drawing a straight line from the position on each predictor scale, a corresponding prognostic point is found on the point scale. The prognostic points for all variables are added up, and the probability of early HCC recurrence is estimated from the bottom line with total points. The performance of this model was quantified with respect to discrimination (27). Discrimination (ie, whether the relative ranking of individual predictions of subsequent early recurrence is in the correct order) was quantified with the area under the receiver operating characteristic curve (AUC). We also used

Radiology: Volume 000: Number 0—   2015  n  radiology.rsna.org

the bootstrapping method (200 repetitions) to obtain relatively unbiased estimates. To compare the discriminative ability between the preoperative prediction model, which used MR imaging features, and the postoperative prediction model, which used pathologic findings, we performed pairwise comparison of the AUCs for the two prediction models. Interobserver or intermethod agreement about the presence of MR imaging features was expressed with the Cohen k coefficient. A k statistic of 0.8– 1 was considered indicative of excellent agreement; 0.6–0.79, good agreement; 0.4–0.59, moderate agreement; 0.2– 0.39, fair agreement; and 0–0.19, poor agreement (28). The validity of size measurements obtained with MR imaging against those obtained with histologic examination was analyzed with the one-way model intraclass correlation coefficient. The intraclass correlation coefficient was classified as poor (,0.21), fair (0.21–0.4), moderate (0.41–0.6), good (0.61–0.8), or excellent (0.81–1) (29). Two-sided P , .05 was considered indicative of a statistically significant difference. All analyses were performed by using software (version 9.2; SAS Institute, Cary, NC). 5

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

Results The characteristics of patients in the training and validation sets are summarized in Table 2. The early recurrence rate was not significantly different between the two patient groups (29.9% for the training set vs 37% for the validation set, P = .318). There were no significant differences in mean age, male-to-female ratio, Child-Pugh class, cause of liver disease, and serum levels of the tumor markers (AFP and PIVKAII) between patients in the training and validation sets. All pathologic findings (histologic differentiation, microvessel or gross vascular invasion, satellite nodule, resection margin, and serosal invasion) showed no significant differences in frequency between the two groups. Among MR imaging features, the proportion of patients with rim enhancement (11.8% in the training set vs 14.8% in the validation set) and coexisting satellite nodules (7% in the training set vs 11.1% in the validation set) were also similar in both sets. The frequency of peritumoral parenchymal enhancement, however, was significantly greater in the training set than in the validation set (35.8% vs 24.7%, respectively; P = .006). Logistic regression was performed with preoperatively assessable variables (tumor markers and MR imaging features) to establish a model for predicting early HCC recurrence. Univariate analysis in the training set (n = 187) showed the following parameters to be significantly associated with early recurrence: tumor size, presence of rim enhancement, peritumoral parenchymal enhancement, peritumoral hypointensity on hepatobiliary images, irregular tumor margin, satellite nodules, gross vascular invasion, and PIVKA-II level greater than 400 mAU/mL (vs 9 mAU/ mL). The following parameters were not significantly related to prognosis: tumor capsule, intratumoral fat at MR imaging, tumor-to-liver signal intensity ratio on hepatobiliary and/or diffusionweighted images, ADC, and serum level of AFP (Table 3). When a multiple logistic regression model was fitted to all variables that showed significant 6

An et al

Table 2 Patient Characteristics in the Training and Validation Sets Parameter Mean age (y) M/F ratio Child-Pugh class  A  B Cause of liver disease  HBV  HCV  Alcoholism  Other Early recurrence  Yes  No AFP level†   9 ng/mL   .9 and 400 ng/mL   .400 ng/mL  NA PIVKA-II level†   40 mAU/mL   .40 and 400 mAU/mL   .400 mAU/mL  NA Qualitative MR imaging features   Rim enhancement   Present   Absent   Peritumoral enhancement   Present   Absent  Capsule   Present   Absent   Peritumoral low signal intensity on HBP images   Present   Absent   Irregular tumor margin   Present   Absent   Satellite nodule   Present   Absent   Intratumoral fat   Present   Absent   Gross vascular invasion   Present   Absent Quantitative MR imaging features§    Tumor size (cm)

Training Set (n = 187)*

Validation Set (n = 81)*

58.7 6 10.1 286:72

57.2 6 10.4 65:16

182 (97.3) 5 (2.7)

79 (98) 2 (2.5)

164 (87.8) 7 (3.7) 7 (3.7) 9 (4.8)

68 (84) 6 (7.4) 4 (4.9) 3 (3.7)

56 (29.9) 131 (70.1)

30 (37) 51 (63)

65 (41.7) 60 (38.5) 31 (19.8) 25

27 (39) 26 (37) 17 (24) 11

56 (40.9) 48 (35) 33 (24.1) 44

28 (43) 24 (37) 13 (20) 16

22 (11.8) 165 (88.2)

12 (15) 69 (85)

67 (35.8) 120 (64.2)

20 (25) 61 (75)

60 (32.1) 127 (67.9)

… …

43 (23) 144 (77)

… …

P Value .27 .936 .749

Q10

.571

.318

.749

.811

.549

.006‡

Q11

90 (48.1) 97 (51.9)

… … .312

13 (7) 174 (93)

9 (11) 72 (89)

45 (24.1) 142 (75.9

… …

10 (5.3) 177 (94.7)

… …

3.2 (0.8–8.9)

2.8 (1.1–12.6)

.171

(Table 2 continues)

radiology.rsna.org  n Radiology: Volume 000: Number 0—   2015

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

Table 2 (continued) Patient Characteristics in the Training and Validation Sets Parameter    SIR on HBP images    SIR on DW images    ADC (31023 mm2/sec) Histologic features   Histologic differentiation    Well differentiated   Moderately differentiated   Poorly differentiated   Microvessel invasion   Present   Absent   Gross vascular invasion   Present   Absent   Satellite nodule   Present   Absent   Positive resection margin   Present   Absent   Serosal invasion   Present   Absent

Training Set (n = 187)*

Validation Set (n = 81)*

0.4 (0.2–3.3) 2.8 (0.9–9.3) 0.9 (0.2–1.5)

… … …

12 (6.4) 77 (41.2) 98 (52.4)

6 (7.4) 37 (45.7) 38 (46.9)

P Value

.709

.087 106 (56.7) 81 (43.3)

36 (44) 45 (55)

6 (3.2) 181 (96.8)

2 (2.5) 79 (97)

11 (5.9) 176 (94.1)

7 (8.6) 74 (91)

3 (1.6) 184 (98.4)

1 (1.2) 80 (99)

11 (5.9) 176 (94.1)

8 (9.9) 73 (90)

Q12

.949

.431

.749

.362

Note.—DW = diffusion-weighted, HBP = hepatobiliary phase, HBV = hepatitis B virus, HCV = hepatitis C virus, NA = not assessable, SIR = tumor-to-liver signal intensity ratio. * Except where indicated, data are numbers of patients, with percentages in parentheses. †

Percentages were calculated after excluding the nonassessable (missing) cases.

‡

Statistically significant with the x2 test.

§

Data are medians. Numbers in parentheses are the range.

association with early recurrence at univariate analysis, the following risk factors attained statistical significance: rim enhancement (OR = 3.83; 95% CI: 1.39, 10.52; P = .01), peritumoral parenchymal enhancement (OR = 2.64; 95% CI: 1.27, 5.46; P , .009), satellite nodules (OR = 4.07; 95% CI: 1.09, 15.21; P = .037), and tumor size (OR = 1.66; 95% CI: 1.31, 2.09; P , .001). There was no multicollinearity between any of these independent variables. A multivariate model to predict early HCC recurrence was developed by using the independent predictors (rim enhancement, peritumoral parenchymal enhancement, satellite nodule, and tumor size). A regression coefficient– based nomogram was constructed from

these predictors (Fig 3). The discrimination accuracy determined with the AUC was comparable (P = .804) between the training set (AUC, 0.788) and validation set (AUC, 0.783), which suggests that the discrimination was as good in the validation set as in the training set (Fig 4). When pathologic findings were used as variables, results of univariate and multivariate logistic regression analyses showed that larger tumor size (OR = 1.45; 95% CI: 1.16, 1.81; P = .001), microvascular invasion (OR = 2.87; 95% CI: 1.31, 6.33; P = .009), and satellite nodules (OR = 4.62; 95% CI: 1.12, 32.42; P = .029) were independently associated with increased risk of early HCC recurrence in the training set (Table 4). The relationship

Radiology: Volume 000: Number 0—   2015  n  radiology.rsna.org

An et al

between gross vascular invasion and early recurrence was statistically significant at univariate analysis but not at multivariate analysis. Histologic differentiation, serosal tumor invasion, and a positive resection margin were not significantly related to early HCC recurrence. The discrimination accuracy calculated from pathologic findings (tumor size, microvascular invasion, and satellite nodules) was comparable with that obtained from MR imaging features in the validation group (AUC, 0.763 vs 0.783; P = .621). Cross-tabulation with the x2 or Fisher exact test revealed that microvascular invasion was significantly associated with rim enhancement, peritumoral parenchymal enhancement, capsule, peritumoral hypointensity on the hepatobiliary phase images, and irregular tumor margin (Table E1 [online]). However, only rim enhancement (OR = 4.52; 95% CI: 1.21, 16.82; P = .025) and peritumoral parenchymal enhancement (OR = 2.61; 95% CI: 1.16, 5.87; P = .021) were independent predictors for the presence of microvascular invasion at multivariate logistic regression analysis (Table E2 [online]). Interobserver agreement about the presence of the three MR imaging features that were independently associated with early recurrence was good (k = 0.72 for rim enhancement, 0.738 for peritumoral parenchymal enhancement, and 0.721 for satellite nodules). Intermethod agreement between MR images and pathology reports was excellent with regard to tumor size (intraclass correlation coefficient = 0.93, with a narrow 95% CI [0.91, 0.95]).

Discussion The results of this study demonstrate that a prediction model derived from rim arterial enhancement, peritumoral parenchymal enhancement, satellite nodule, and tumor size determined with preoperative MR imaging features is a useful tool for estimating the probability of early recurrence after resection of a single HCC in patients with preserved liver function. Preoperative identification of patients at high risk of 7

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

Table 3 Univariate and Multivariate Analysis of Preoperative MR Imaging Findings and Tumor Markers Predictive of Early Recurrence in the Training Set Univariate Analysis Parameter Rim enhancement Peritumoral enhancement Capsule Peritumoral hypointensity   on HBP images Irregular tumor margin Satellite nodule Intratumoral fat Gross vascular invasion Tumor size SIR on HBP images SIR on DW images ADC (31023 mm2/sec) AFP level   9 ng/mL†   .9 and 400 ng/mL   .400 ng/mL PIVKA-II level   40 mAU/mL†   .40 and 400 mAU/mL   .400 mAU/mL

OR

Multivariate Analysis

95% CI

P Value

OR

95% CI

P Value

3.3 3.27 0.7 2.97

1.33, 8.18 1.71, 6.28 0.35, 1.4 1.46, 6.04

.009* ,.001* .311 .003*

3.83 2.64

1.39, 10.52 1.27, 5.46

.01* .009*

1.11

0.43, 2.88

.831

2.07 4.2 0.6 10.75 1.7 0.41 1.02 1.01

1.09, 3.92 1.31, 13.48 0.27, 1.31 2.2, 52.43 1.36, 2.11 0.1, 1.72 0.82, 1.25 0.98, 1.03

.026* .016* .197 .003* ,.001* .222 .878 .889

1.13 4.07

0.52, 2.44 1.09, 15.21

.765 .037*

4.22 1.66 … … …

0.74, 24.06 1.31, 2.09 … … …

.105 ,.001* … … …

… 2.49 1.91

… 0.89, 5.54 0.72, 5.01

… .079 .192

… … …

… … …

… … …

… 1.51 3.06

… 0.12, 42.09 1.19, 7.79

… .133 .02*

… … 1.08

… … 0.35, 3.98

… … 0.791

Note.—CI = confidence interval, DW = diffusion-weighted, HBP = hepatobiliary phase, OR = odds ratio, SIR = tumor-to-liver signal intensity ratio. * Statistically significant results from logistic regression analysis. †

Used as the reference category.

recurrence after HCC resection is important because there is currently no effective adjuvant or neoadjuvant therapy (30). In this regard, we believe that our results demonstrated that a prediction model that uses preoperative MR imaging has the potential to preoperatively identify high-risk patients, for whom other treatment modalities such as liver transplantation or a wider extent of resection can be considered. Herein, we presented a nomogram created from multivariable logistic regression analysis as a tool for individualized risk estimation. As opposed to the equation of an underlying logistic model, which requires complicated calculations, a nomogram enables users to graphically compute the numerical probability of a clinical event without resorting to a calculator or a computer. Thus, this user-friendly graphical tool 8

can be useful for assisting physicians in clinical decision-making and preoperative discussion with patients and has gained popularity among physicians and patients themselves (27). Our prediction models based on MR imaging features and pathologic findings were comparable in terms of discriminative ability determined by using AUCs. The explanatory variables included in the pathology-based model were tumor size, microvascular invasion, and satellite nodule. These variables are key prognostic determinants used in the American Joint Committee on Cancer staging system, which is reportedly superior to other staging systems in patients undergoing curative surgery for HCC (31–33). Among the three predictors of the pathology-based model, microvascular invasion cannot be reliably evaluated with MR imaging,

An et al

but the other two variables (tumor size and satellite nodule) were also included in the MR imaging–based model and showed similar ORs. In the MR imaging–based model, two MR imaging findings (rim enhancement and peritumoral parenchymal enhancement) were included in place of microvascular invasion. These observations suggest that these two MR imaging findings could be as helpful for predicting early recurrence as pathologically determined microvascular invasion. Peritumoral parenchymal enhancement and rim enhancement, which were included in our prediction model, were also independent risk factors for the presence of microvascular invasion. These results agree with those from previous studies in which peritumoral parenchymal enhancement or tumorous arterioportal shunt was associated with microvascular invasion (16,17,34). The rim enhancement pattern has been also suggested to indicate rapid progression, poor differentiation, and worse prognosis (22,25,35), possibly by reflecting the absence of a capsule, infiltrative growth, and the presence of microvascular invasion as well as rapid growth with central necrosis. Peritumoral low signal intensity and irregular tumor margin on hepatobiliary images were excluded from the independent risk factors in the multivariate analysis, although they were significant factors at univariate analysis. Both MR imaging findings were also not significantly correlated with the presence of microvascular invasion at multivariate analysis, which indicates that rim enhancement and peritumoral parenchymal enhancement are better predictors of both microvascular invasion and early recurrence than peritumoral low signal intensity and irregular tumor margin evaluated on hepatobiliary images. A counterintuitive result of our study is that gross vascular invasion was not an independent risk factor of early recurrence at multivariate analysis, although it was significantly associated with early HCC recurrence at univariate analysis. Although all but one case with gross vascular invasion suspected

radiology.rsna.org  n Radiology: Volume 000: Number 0—   2015

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

Figure 4

Figure 4:  Receiver operating characteristic curves obtained with prediction model in training and validation data sets.

Table 4 Univariate and Multivariate Analysis of Postoperative Pathologic Findings Predictive of Early Recurrence in the Training Set Univariate Analysis Parameter Histologic differentiation   Well differentiated*   Moderately differentiated   Poorly differentiated Tumor size Microvessel invasion Gross vascular invasion Satellite nodule Positive resection margin Serosal invasion

Multivariate Analysis

OR

95% CI

P Value

OR

95% CI

P Value

… 2.09 5.06 1.56 4.49 12.65 12.56 4.78 2.98

… 0.24, 18.01 0.61, 42.09 1.28, 1.91 2.13, 9.46 1.44, 110.92 2.55, 58.8 0.42, 53.81 0.87, 10.19

… .499 .133 ,.001† ,.001† .022† .002† .206 .083

… … … 1.45 2.87 4.13 4.62 … …

… … … 1.16, 1.81 1.31, 6.33 0.38, 45.56 1.12, 32.42 … …

… … … .001† .009† .247 .029† … …

* Reference value. †

Statistically significant results from logistic regression analysis.

at MR imaging proved to have microvascular invasion at pathologic examination (seven of eight cases, 88%), statistical analysis did not show a significant relationship between the two variables. These findings are probably because of the highly selected patients included in this study. All of our patients were candidates for curative

resection, in whom gross vascular invasion may have been less frequent (5.3% in the present study) and less severe (invasion of small branches rather than main blood vessels) than is seen in advanced HCCs, and this may have led to no additional effect of the presence of gross vascular invasion on the risk of early recurrence.

Radiology: Volume 000: Number 0—   2015  n  radiology.rsna.org

An et al

Other quantitative and qualitative imaging features reported to be related to prognosis, such as tumor-to-liver signal intensity on hepatobiliary images, encapsulation, and the presence of intratumoral fat, did not show significant associations with early recurrence in our study population. This discordance may be attributed to a relatively small number of tumors that showed high signal intensity on hepatobiliary phase images. This explanation can also be applied to the discordant results on encapsulation and intratumoral fat. Previous studies that investigated these imaging features tended to have a higher proportion of tumors with intratumoral fat and encapsulation than in our study (36– 38). A few previous studies consistently found that higher signal intensity on diffusion-weighted images was associated with less histologic differentiation, and a possible relationship between ADCs and histologic differentiation and microvascular invasion has been suggested (39–42). However, in our study, neither histologic differentiation nor signal intensity or ADC on diffusion-weighted images was a significant predictor of early HCC recurrence. In addition to MR imaging findings, information related to tumor markers is also obtainable before surgery and can potentially be used to preoperatively predict the prognosis after curative resection of HCC. However, previous studies reported conflicting results regarding the association between tumor markers and early recurrence (15,43,44). In our study, neither AFP nor PIVKA-II was a significant predictor of early recurrence. The discrepancy between the results may be attributed to the heterogeneity of patient populations evaluated and the different cutoff levels of tumor markers used in the other studies. One limitation of this study is its retrospective nature. However, we believe that we could compensate for this limitation by applying our model derived from the training set to the validation set to increase its reliability. Second, our study is from a single-center experience. Therefore, our results may need further validation in different 9

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

institutions with use of prospective studies. In addition, the frequency of peritumoral parenchymal enhancement, one of the independent predictors of early recurrence in the training cohort, was significantly higher in the training set than in the validation set, which resulted in some heterogeneity between the two data sets. Inclusion of a larger number of patients would be necessary for further validation. Another limitation is that our prediction model assumes that the effect of tumor size on the probability of recurrence is linear, which may not reflect the true association. It is also noteworthy that we included only the patients with a single tumor regardless of satellite nodules, partly because only patients with a single HCC are recommended for surgical resection according to the Barcelona Clinical Liver Cancer system, which is one of the current widely used HCC staging systems. Therefore, our model cannot be directly translated to patients who undergo surgical resection of more than one tumor. A future study to investigate the utility of our prediction model in patients with multiple HCCs may be worthwhile. Last, we used gadoxetic acid, a hepatocyte-specific contrast material, for all MR images. However, our final prediction model contains only MR findings that can be evaluated without hepatobiliary phase images, which means that the probability of early recurrence could be estimated with our prediction model even when extracellular contrast material is used. A further study is warranted to determine whether our results can be reproduced with extracellular contrast material. In conclusion, the results of our study showed that rim arterial enhancement, peritumoral parenchymal enhancement, larger tumor size, and the presence of satellite nodules are independent predictors of early recurrence (,2 years) after curative resection of a single HCC. The nomogram developed on the basis of these predictors can be used preoperatively to estimate the risk of early recurrence. Disclosures of Conflicts of Interest: C.A. disclosed no relevant relationships. D.W.K. disclosed no relevant relationships. Y.N.P. disclosed

10

no relevant relationships. Y.E.C. disclosed no relevant relationships. H.R. disclosed no relevant relationships. M.J.K. disclosed no relevant relationships.

References 1. Roayaie S, Obeidat K, Sposito C, et al. Resection of hepatocellular cancer 2 cm: results from two Western centers. Hepatology 2013;57(4):1426–1435. 2. Bruix J, Sherman M; American Association for the Study of Liver Diseases. Management of hepatocellular carcinoma: an update. Hepatology 2011;53(3):1020–1022. 3. Harada T, Shigemura T, Kodama S, Higuchi T, Ikeda S, Okazaki M. Hepatic resection is not enough for hepatocellular carcinoma: a follow-up study of 92 patients. J Clin Gastroenterol 1992;14(3):245–250. 4. Nagasue N, Uchida M, Makino Y, et al. Incidence and factors associated with intrahepatic recurrence following resection of hepatocellular carcinoma. Gastroenterology 1993;105(2):488–494. 5. Shah SA, Cleary SP, Wei AC, et al. Recurrence after liver resection for hepatocellular carcinoma: risk factors, treatment, and outcomes. Surgery 2007;141(3):330–339. 6. Tung-Ping Poon R, Fan ST, Wong J. Risk factors, prevention, and management of postoperative recurrence after resection of hepatocellular carcinoma. Ann Surg 2000;232(1):10–24. 7. Lee JI, Lee JW, Kim YS, Choi YA, Jeon YS, Cho SG. Analysis of survival in very early hepatocellular carcinoma after resection. J Clin Gastroenterol 2011;45(4):366–371. 8. Choi GH, Kim DH, Kang CM, et al. Prognostic factors and optimal treatment strategy for intrahepatic nodular recurrence after curative resection of hepatocellular carcinoma. Ann Surg Oncol 2008;15(2):618–629. 9. Portolani N, Coniglio A, Ghidoni S, et al. Early and late recurrence after liver resection for hepatocellular carcinoma: prognostic and therapeutic implications. Ann Surg 2006;243(2):229–235. 10. Poon RT, Fan ST, Ng IO, Lo CM, Liu CL, Wong J. Different risk factors and prognosis for early and late intrahepatic recurrence after resection of hepatocellular carcinoma. Cancer 2000;89(3):500–507.

An et al

phase intrahepatic recurrence of hepatocellular carcinoma after hepatectomy. J Hepatol 2003;38(2):200–207. 13. Kamiyama T, Nakanishi K, Yokoo H, et al. Analysis of the risk factors for early death due to disease recurrence or progression within 1 year after hepatectomy in patients with hepatocellular carcinoma. World J Surg Oncol 2012;10:107. 14. Park JH, Koh KC, Choi MS, et al. Analysis of risk factors associated with early multinodular recurrences after hepatic resection for hepatocellular carcinoma. Am J Surg 2006;192(1):29–33. 15. Cucchetti A, Piscaglia F, Caturelli E, et al. Comparison of recurrence of hepatocellular carcinoma after resection in patients with cirrhosis to its occurrence in a surveilled cirrhotic population. Ann Surg Oncol 2009;16(2):413–422. 16. Miyata R, Tanimoto A, Wakabayashi G, et al. Accuracy of preoperative prediction of microinvasion of portal vein in hepatocellular carcinoma using superparamagnetic iron oxide–enhanced magnetic resonance imaging and computed tomography during hepatic angiography. J Gastroenterol 2006;41(10):987–995. 17. Kim H, Park MS, Choi JY, et al. Can microvessel invasion of hepatocellular carcinoma be predicted by pre-operative MRI? Eur Radiol 2009;19(7):1744–1751. 18. Ariizumi S, Kitagawa K, Kotera Y, et al. A non-smooth tumor margin in the hepatobiliary phase of gadoxetic acid disodium (Gd-EOB-DTPA)–enhanced magnetic resonance imaging predicts microscopic portal vein invasion, intrahepatic metastasis, and early recurrence after hepatectomy in patients with hepatocellular carcinoma. J Hepatobiliary Pancreat Sci 2011;18(4):575–585. 19. Chandarana H, Robinson E, Hajdu CH, Drozhinin L, Babb JS, Taouli B. Microvascular invasion in hepatocellular carcinoma: is it predictable with pretransplant MRI? AJR Am J Roentgenol 2011;196(5):1083–1089. 20. Griffin N, Addley H, Sala E, et al. Vascu lar invasion in hepatocellular carcinoma: is there a correlation with MRI? Br J Radiol 2012;85(1014):736–744.

11. Shimada M, Takenaka K, Gion T, et al. Prognosis of recurrent hepatocellular carcinoma: a 10-year surgical experience in Japan. Gastroenterology 1996;111(3):720–726.

21. Kim KA, Kim MJ, Jeon HM, et al. Prediction of microvascular invasion of hepatocellular carcinoma: usefulness of peritumoral hypointensity seen on gadoxetate disodium-enhanced hepatobiliary phase images. J Magn Reson Imaging 2012;35(3):629–634.

12. Imamura H, Matsuyama Y, Tanaka E, et al. Risk factors contributing to early and late

22. Kierans AS, Leonardou P, Hayashi P, et al. MRI findings of rapidly progressive hepa-

radiology.rsna.org  n Radiology: Volume 000: Number 0—   2015

GASTROINTESTINAL IMAGING: Prediction of Early Recurrence of Single Hepatocellular Carcinoma

An et al

tocellular carcinoma. Magn Reson Imaging 2010;28(6):790–796.

patocellular carcinoma. Cochrane Database Syst Rev 2009 (1):CD001199.

carcinoma. J Surg Oncol 2012;105(1):85– 90.

23. Kawamura Y, Ikeda K, Seko Y, et al. Heterogeneous type 4 enhancement of hepatocellular carcinoma on dynamic CT is associated with tumor recurrence after radiofrequency ablation. AJR Am J Roentgenol 2011;197(4):W665–W673.

31. Sirivatanauksorn Y, Tovikkai C. Comparison of staging systems of hepatocellular carcinoma. HPB Surg 2011;2011:818217.

39. An C, Park MS, Jeon HM, et al. Prediction of the histopathological grade of hepatocellular carcinoma using qualitative diffusionweighted, dynamic, and hepatobiliary phase MRI. Eur Radiol 2012;22(8):1701–1708.

24. Choi JW, Lee JM, Kim SJ, et al. Hepatocellular carcinoma: imaging patterns on gadoxetic acid–enhanced MR images and their value as an imaging biomarker. Radiology 2013;267(3):776–786. 25. Kawamura Y, Ikeda K, Hirakawa M, et al. New classification of dynamic computed tomography images predictive of malignant characteristics of hepatocellular carcinoma. Hepatol Res 2010;40(10):1006–1014. 26. Roayaie S, Blume IN, Thung SN, et al. A system of classifying microvascular invasion to predict outcome after resection in patients with hepatocellular carcinoma. Gastroenterology 2009;137(3):850–855. 27. Iasonos A, Schrag D, Raj GV, Panageas KS. How to build and interpret a nomogram for cancer prognosis. J Clin Oncol 2008;26(8):1364–1370. 28. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33(1):159–174.

32. Nathan H, Mentha G, Marques HP, et al. Comparative performances of staging systems for early hepatocellular carcinoma. HPB (Oxford) 2009;11(5):382–390. 33. Vauthey JN, Dixon E, Abdalla EK, et al. Pretreatment assessment of hepatocellular carcinoma: expert consensus statement. HPB (Oxford) 2010;12(5):289–299. 34. Nishie A, Yoshimitsu K, Asayama Y, et al. Radiologic detectability of minute portal venous invasion in hepatocellular carcinoma. AJR Am J Roentgenol 2008;190(1):81–87. 35. Honda H, Hayashi T, Yoshida K, et al. Hepatocellular carcinoma with sarcomatous change: characteristic findings of two-phased incremental CT. Abdom Imaging 1996;21(1): 37–40. 36. Siripongsakun S, Lee JK, Raman SS, Tong MJ, Sayre J, Lu DS. MRI detection of intratumoral fat in hepatocellular carcinoma: potential biomarker for a more favorable prognosis. AJR Am J Roentgenol 2012;199(5):1018–1025.

29. McGraw OK, Wong SP. Forming inferences about some intraclass correlation coefficients. Psychol Methods 1996;1(1):30–46.

37. Woodall CE, Scoggins CR, Loehle J, Ravindra KV, McMasters KM, Martin RC. Hepatic imaging characteristics predict overall survival in hepatocellular carcinoma. Ann Surg Oncol 2007;14(10):2824–2830.

30. Samuel M, Chow PK, Chan Shih-Yen E, Machin D, Soo KC. Neoadjuvant and adjuvant therapy for surgical resection of he-

38. Wu TH, Yu MC, Chen TC, et al. Encapsulation is a significant prognostic factor for better outcome in large hepatocellular

Radiology: Volume 000: Number 0—   2015  n  radiology.rsna.org

40. Heo SH, Jeong YY, Shin SS, et al. Apparent diffusion coefficient value of diffusionweighted imaging for hepatocellular carcinoma: correlation with the histologic differentiation and the expression of vascular endothelial growth factor. Korean J Radiol 2010;11(3):295–303. 41. Nasu K, Kuroki Y, Tsukamoto T, Nakajima H, Mori K, Minami M. Diffusion-weighted imaging of surgically resected hepatocellular carcinoma: imaging characteristics and relationship among signal intensity, apparent diffusion coefficient, and histopathologic grade. AJR Am J Roentgenol 2009;193(2):438–444. 42. Nishie A, Tajima T, Asayama Y, et al. Diagnostic performance of apparent diffusion coefficient for predicting histological grade of hepatocellular carcinoma. Eur J Radiol 2011;80(2):e29–e33. 43. Shah SA, Greig PD, Gallinger S, et al. Factors associated with early recurrence after resection for hepatocellular carcinoma and outcomes. J Am Coll Surg 2006;202(2):275– 283. 44. Regimbeau JM, Abdalla EK, Vauthey JN, et al. Risk factors for early death due to recurrence after liver resection for hepatocellular carcinoma: results of a multicenter study. J Surg Oncol 2004;85(1):36–41.

11