Hepatocellular carcinoma (HCC) is the most common primary hepatic malignancy and is the third leading cause of cancer-related deaths worldwide.1 Although the treatment of HCC is evolving, hepatic resection remains the treatment of choice for many patients.2 However, the results of hepatic resection for HCC are still unsatisfactory because of tumor recurrence rates as high as 70% within 5 years.3 The success of curative resection depends on tumor characteristics such as the size, number, histological grade, and presence of vascular invasion.4, 5 Among these characteristics, vascular invasion is regarded as the most important factor affecting the success of curative resection.6-8 Also, vascular invasion has been previously associated with poor outcomes after liver transplantation.9 Vascular invasion can be divided into macrovascular invasion and microvascular invasion (MVI). Macrovascular invasion in major vessels is known to be a marker of poor outcomes after liver transplantation for HCC and is regarded as a contraindication to liver transplantation. Although the significance of MVI as a predictor of poor posttransplant outcomes is still controversial, the majority of studies favor the idea of a relationship between MVI and poor posttransplant outcomes.9 Although macrovascular invasion can be detected with imaging studies, the preoperative prediction of MVI remains difficult.7, 8, 10-16
Diffusion-weighted imaging (DWI) is an emerging technique in hepatic magnetic resonance imaging (MRI).17 DWI may improve the detection of HCC18-20 and provide additional information that is useful for differentiating HCC from dysplastic nodules.21 DWI has also been applied to the preoperative assessment of the HCC histological grade,22-25 which is known to be a predictor of MVI.26 However, the relationship between the DWI features of HCC and the presence of MVI is unknown. In this study, we investigated whether DWI could be useful in predicting MVI in HCC during the preoperative evaluation.
ADC, apparent diffusion coefficient; CI, confidence interval; DW, diffusion-weighted; DWI, diffusion-weighted imaging; HCC, hepatocellular carcinoma; MR, magnetic resonance; MRI, magnetic resonance imaging; MVI, microvascular invasion; N/A, not applicable; ROI, region of interest; SI, signal intensity; SIlesion, lesion signal intensity; SIliver, liver signal intensity; TE, effective echo time; TR, repetition time.
PATIENTS AND METHODS
This retrospective study was approved by the institutional review board of our hospital. The requirement of informed consent was waived.
We reviewed information from our radiology database (June 2009 to February 2010) and retrieved data on 75 lesions from 73 patients who underwent hepatic MRI (including DWI) and subsequent surgical resection for the diagnosis of HCC. Magnetic resonance (MR) examinations had been performed for the preoperative evaluation of known or suspected HCC. We excluded 8 lesions from 8 patients who received transarterial chemoembolization between MRI and the operation. No patient had previously undergone any preoperative treatment other than transarterial chemoembolization.
In all, 67 HCCs from 65 patients (54 men and 11 women with an age range of 35-75 years and a mean age of 56.0 years) were included in this study. Two patients had 2 HCCs each; the remaining 63 patients had 1 lesion each. Sixty of the 65 patients (92.3%) had hepatitis B, and 1 (1.5%) had hepatitis C. One patient had liver cirrhosis associated with chronic alcoholic hepatitis, and another had biliary liver cirrhosis associated with a previous Clonorchis sinensis infection. Two patients (3.1%) exhibited no evidence of chronic liver disease. The mean interval between MRI and surgery was 12.7 days (range = 1-45 days, median = 9 days).
All patients were examined with a 3.0-T MR system (Magnetom Trio a Tim, Syngo MR B15, Siemens Medical Solutions, Erlangen, Germany). All images were obtained in the transverse plane with a single dedicated body phased-array coil anterior and a spine-array coil posterior. Routine liver MRI was performed with the following sequences: 2 breath-hold, T1-weighted, spoiled gradient-recalled echo, in-phase sequences [repetition time (TR)/effective echo time (TE) = 140 ms/1.22 ms] and an out-of-phase sequence (150 ms/2.5 ms) with a flip angle of 65°, 1 acquired signal, a 256 × 192 matrix, a 7-mm slice thickness, and a 0.7-mm gap; a navigator-triggered, T2-weighted, turbo spin-echo sequence with a TR range of 3300 to 4900 ms, a TE of 73 ms, an echo train length of 14, 1 acquired signal, a 320 × 179 matrix, superior and inferior spatial presaturation and chemical fat saturation, a 4-mm slice thickness, and a 1-mm gap; and a breath-hold, heavily T2-weighted, half-Fourier acquisition, turbo spin-echo sequence with a TE of 150 ms, a 320 × 179 matrix, a 4-mm slice thickness, and a 1-mm gap. For dynamic MRI, gadoxetic acid (Primovist, Bayer Schering; 0.025 mmol/kg) was injected as a rapid bolus, and it was immediately followed by a 30-mL saline flush through a power injector at a rate of 1 or 2 mL/second. A 3-dimensional, spoiled gradient-recalled echo sequence with chemical selective fat suppression was performed before the injection of the intravenous contrast agent. Contrast-enhanced images were obtained 20 to 30, 60 to 70, 90 to 100, and 120 to 150 seconds after the injection. Hepatobiliary phase images were obtained 15 to 20 minutes after the contrast injection. The MR parameters included a TR/TE ratio of 3.3 ms/1.16 ms, a flip angle of 13°, a 256 × 192 matrix, 1 acquired signal, and a 2.5-mm slice thickness with an intersection gap of zero.
Diffusion-Weighted (DW) MRI
Free-breathing DWI was performed with a single-shot, echo-planar sequence with motion-probing gradients in 3 directions during the interval between dynamic imaging and hepatobiliary phase imaging with the following parameters: b values of 50, 400, and 800 second/mm2; a TR/TE ratio of 1400 ms/74 ms; 2 acquired signals; a 192 × 153 matrix; an 8-mm slice thickness; and a 2-minute acquisition time. The apparent diffusion coefficients (ADCs) for each DW image were automatically calculated with the MR system and were displayed as corresponding ADC maps.
The image analysis was performed by a primary investigator (Y.J.S.) who was blinded to the histological grades of the lesions and the presence of MVI. The tumor size, which was defined as the maximum diameter, was measured on hepatobiliary phase images. For the qualitative analysis, the signal intensity (SI) of each lesion was visually assessed and was classified as isointense or hyperintense with respect to the intensity of the adjacent hepatic parenchyma on DW images with a high b value (800 second/mm2).
For the quantitative analysis, regions of interest (ROIs) were drawn in the HCCs and hepatic parenchyma at a workstation (Centricity 2.0, GE Healthcare, United States) so that we could measure the SI values on DW images with b values of 50, 400, and 800 second/mm2. The area of the ROI in each HCC was set to include the entire lesion, did not exclude components with different attenuation values, and covered an area of greater than 20 mm2 (mean = 152.5 mm2, range = 21-519.9 mm2). The ROI in each liver was placed in the posterior right hepatic lobe (specifically at the level of the main portal vein and its right branches), and vessels and artifacts were excluded. The areas of the ROI in the liver were greater than 100 mm2 (mean = 248.4 mm2, range = 104.2-798.3 mm2). ROIs were manually positioned to ensure their identical placement on DW images with b values of 50, 400, and 800 second/mm2. The lesion-to-liver SI ratios were calculated with the following equation:
where SIlesion is the lesion signal intensity and SIliver is the liver signal intensity. ADC values were measured through the placement of ROIs in the HCCs and hepatic parenchyma on ADC maps. To ensure the identical placement of ROIs on DW images and ADC maps, ROIs were carefully positioned in the same regions on the corresponding ADC maps. Two lesions were invisible on DW images, and 12 lesions were considered invisible on ADC maps. When the lesions were invisible on DW images or ADC maps, the ADC measurements were performed in the same area in which the ROIs were placed for other sequences by the visual correlation of the image sets.
We used original pathology reports for the analysis of histopathological features. The presence of MVI in HCCs was evaluated on the basis of pathological reports of surgical specimens. At our institution, whole HCC specimens were retrieved, and all tissue sections were systematically sliced into thin slices (5 mm). The slices were fixed in a 10% buffered formaldehyde solution and paraffinized. All lesions that were suspected to be HCC or dysplastic nodules were photographed and carefully documented for their shape, color, texture, segmental location, and size (long and short axis diameters). In accordance with the guidelines developed by the Korean Liver Cancer Study Group for the Study of Primary Liver Cancer, the following histopathological factors were assessed for each tumor: nuclear grade, growth type, histological type, cell type, gross invasion and MVI, fibrous capsule formation, and capsular infiltration.27 MVI was defined as a tumor within a vascular space lined by endothelium that was visible only on microscopy.5 The HCC tumor grades were classified according to the Edmondson-Steiner nuclear grading system.28 When different tumor grades were found within the same tumor, the predominant grade was used as the tumor grade.
We used the Student t test to determine the relationships between continuous variables (eg, age, tumor size, mean SI ratio, and mean ADC value) and the presence of MVI. We used the Mann-Whitney U test to determine the relationships between serum alpha-fetoprotein and des-gamma-carboxy prothrombin levels and the presence of MVI. The chi-square test was performed to compare qualitative SI, sex, multiplicity, and histological grades with the presence of MVI. We analyzed the proportion of patients with HCC within the Milan criteria (a single lesion up to 5 cm). A logistic regression analysis was used to assess individual clinicopathological and MR findings for the prediction of MVI. Variables with P < 0.05 in a univariate logistic regression analysis were selected as variables for a multiple logistic regression analysis. For the prediction of MVI, receiver operating characteristic curves were drawn to determine the best cutoff values for the continuous variables that were significantly different in the multivariate logistic regression analysis. Odds ratios with 95% confidence intervals (CIs) for predicting MVI were calculated for each significant factor. The diagnostic performance of clinicopathological and MR findings for predicting MVI was analyzed separately for HCCs within and beyond the Milan criteria.
In order to determine the effects of histological grades on the results, we performed the Kruskal-Wallis test or the Mann-Whitney test to assess statistically significant differences in the clinical and imaging variables between different HCC grades. The numbers of lesions were used in this statistical analysis. For all analyses, P values ≤ 0.05 were considered significant. Statistical analyses were performed with MedCalc 11.5 (MedCalc Software, Mariakerke, Belgium).
There were 67 HCCs in 65 patients, and MVI was present in 31 lesions (46.3%) in 31 patients. The 2 patients who had 2 HCCs each had no MVI in either lesion. The diameters of the tumors ranged from 7 to 100 mm (mean = 34.06 mm). HCCs with MVI had higher SI ratios for all 3 DWI b values (P < 0.05) and lower ADC values (P < 0.001) than HCCs without MVI (Fig. 1 and Table 1). HCCs with MVI had a significantly larger mean tumor size and a higher incidence of histological grade 3 than HCCs without MVI. Two of the 36 HCCs without MVI (5.6%) were isointense, whereas no HCCs with MVI were isointense (P = 0.54). One of the grade 1 HCCs and 1 of the grade 2 HCCs were isointense, whereas none of the grade 3 HCCs were isointense (P = 0.11). Fifty-one HCCs satisfied the Milan criteria, and 16 HCCs were beyond the Milan criteria. MVI was present in 23 of the 51 HCCs within the Milan criteria (45.1%). MVI was present in 8 of the 16 HCCs beyond the Milan criteria (50.0%).
Table 1. Clinicopathological and DWI Findings Associated With the Prediction of MVI
All Lesions (n = 67)
MVI-Negative Lesions (n = 36)
MVI-Positive Lesions (n = 31)
The data are presented as means and 95% CIs.
†The data were calculated per lesion and are presented as numbers and percentages of lesions.
The data were calculated per patient and are presented as numbers and percentages of patients.
According to the univariate logistic regression analysis, 4 variables—a histological grade of 3, the tumor size, the SI ratio at a b value of 800 second/mm2, and the ADC value—were significant predictors of MVI (P < 0.05). According to the multiple logistic regression analysis, a histological grade of 3 (P = 0.049) and the ADC value (P < 0.001) were independent predictors of MVI. The odds ratio for a histological grade of 3 was 6.96 (95% CI = 1.01-47.7), and the odds ratio for an ADC value ≤ 1.11 × 10−3 mm2/second was 24.5 (95% CI = 4.14-144.8). The best ADC cutoff value for predicting MVI was 1.11 × 10−3 mm2/second with an area under the receiver operating characteristic curve of 0.898 (Figs. 2 and 3). The sensitivity, specificity, and diagnostic accuracy of this ADC value for predicting MVI reached 93.5% (29/31), 72.2% (26/36), and 82.1% (55/67), respectively. With an ADC cutoff value of 0.97 × 10−3 mm2/second, the specificity reached up to 100%, but the sensitivity was 45.2%. With an ADC cutoff value of 1.15 × 10−3 mm2/second, the sensitivity and specificity were 100% and 61.1%, respectively. The diagnostic performance of a histological grade of 3 was not analyzed because this finding is not available preoperatively. According to the multivariate logistic regression analysis performed for HCCs within the Milan criteria, only the ADC value was an independent predictor of MVI (P = 0.001) with an odds ratio of 29.6 (95% CI = 4.27-204.7). Among HCCs beyond the Milan criteria, none of the histopathological and DWI findings were significant independent predictors of MVI in the multivariate analysis (P > 0.05).
When the histological grades and the ADC values were compared (Table 2), ADC values were significantly lower for grade 2 HCCs with MVI versus grade 2 HCCs without MVI. In addition, ADC values tended to be lower for grade 3 HCCs with MVI versus grade 3 HCCs without MVI. The SI ratios were also higher for grade 2 and 3 tumors with MVI versus grade 2 and 3 tumors without MVI, although the results did not show statistical significance.
Table 2. DWI Variables According to the Histological Grade
NOTE: The data are presented as means and 95% CIs.
For comparison according to MVI with the same histological grade.
b = 50 second/mm2
4.22 (−1.64 to 10.07)
b = 400 second/mm2
b = 800 second/mm2
ADC (10−3 mm2/second)
Tumor size (mm)
In this study, we have demonstrated that DWI findings can be useful for the preoperative prediction of MVI. We used both qualitative and quantitative methods for image analyses, and ADC values were calculated with 3 b values (50, 400, and 800 second/mm2); this potentially increased the reliability of our measurements.
Our results have demonstrated that HCCs with MVI have higher SI ratios and lower ADC values than HCCs without MVI (P < 0.05). In comparison with the SI ratios from DW images, which showed a wide range of overlapping between HCCs with MVI and HCCs without MVI, the ADC values had a relatively narrow range of overlapping and had better diagnostic value. A histological grade of 3 and an ADC value ≤ 1.11 × 10−3 mm2/second were independent predictors of MVI. Moreover, the ADC value was found to be an independent predictor of MVI, although this relationship was significant only for HCCs within the Milan criteria. Although MVI in HCC is one of the most important predictors of recurrence and survival after hepatic resection and liver transplantation, it is difficult to preoperatively predict MVI from imaging studies.10-16 Recently, Griffin et al.15 suggested that conventional MR findings are not correlated with MVI. Chandarana et al.16 found that only tumor multifocality is correlated with MVI. Kim et al.29 suggested that peritumoral hypointensity of the hepatobiliary phase on gadoxetic acid–enhanced MRI has high specificity for predicting MVI in HCC. Kornberg et al.13 reported that increased [18F]fludeoxyglucose-fludeoxyglucose uptake on positron emission tomography is predictive of MVI and tumor recurrence after liver transplantation for HCC. However, to our knowledge, there have been no previous attempts to define the relationship between the DWI features of HCC and the presence of MVI. In our study, the sensitivity and specificity reached 93.5% and 72.2%, respectively, with an ADC cutoff value of 1.11 × 10−3 mm2/second.
Previous studies have suggested that clinicopathological variables other than imaging findings, including age (≤65 years), tumor size (>4 cm) on preoperative imaging, histological grade (poorly differentiated), and des-gamma-carboxy prothrombin levels, are significant predictors of MVI.26, 30 In this study, the presence of MVI was correlated with the histological grade, tumor size, SI ratio at a b value of 800 second/mm2, and ADC value for HCC according to the univariate analysis. However, the histological grade and the ADC value were the only independent predictors of MVI according to the multivariate analysis. Because the histological grade of a tumor cannot be recognized preoperatively, DWI findings such as the ADC value could be especially useful for predicting MVI.
The reason that HCCs with MVI had lower ADC values and higher SI ratios on DW images is difficult to explain. DWI and ADC values provide information related to the tissue cellularity and integrity of cellular membranes, as well as microcapillary perfusion by reflecting the molecular diffusion of water and perfusion.31, 32 We, therefore, propose the following possible mechanisms, although they are not based on histopathological analysis. First, HCCs with MVI may have higher cellularity with restricted diffusion than HCCs without MVI, although it is unknown whether the higher cellularity of HCCs with MVI is the cause of or result of MVI. Second, because the ADC values reflect capillary perfusion as well as molecular diffusion, HCCs with MVI may have decreased perfusion. To assess the alteration of capillary perfusion in HCCs with MVI, further studies with intravoxel incoherent motion imaging may be necessary.33
Another potential explanation of our findings is that DWI and histological grades are related, with poorly differentiated HCCs having lower ADC values than well or moderately differentiated HCCs.22, 24 This is a plausible explanation because a tumor's histological grade has been suggested as a predictor of MVI.34 Our study has demonstrated a significant correlation between the ADC value and the histological grade; however, the ADC value was an independent predictor of MVI according to the multivariate analysis even when the histological grade was included in the analysis. Moreover, there have been conflicting results regarding the correlation between the histological grade and ADC.25 Therefore, the fact that HCCs with MVI presented with significantly lower ADC values cannot be explained solely by the relationship between DWI and the histological grade.
We note that our study has several limitations. First, because of its retrospective nature, the possibility of a selection bias cannot be excluded. The sample sizes for the histological groups were diverse, with a small number of grade 1 tumors (n = 6) and a large number of grade 2 tumors (n = 46). However, this was unavoidable because grade 2 is usually the most prevalent form encountered in clinical practice. Second, for the prediction of MVI, we did not compare the relative accuracy of ADC values and other imaging findings that have been described in previous studies16, 29 because our purpose was to determine whether DWI could be useful for the prediction of MVI, not to assess the relative accuracy of DWI and conventional imaging findings. Future studies should compare DWI findings, conventional imaging findings, and a combination of the two. Third, not only was our study performed with single-reader observations, but there is no standard for ADC measurements, the reproducibility of ADC measurements is still being investigated,35-39 and the reproducibility of our results requires external validation. In addition, our study was conducted with a 3.0-T system. A previous study found that the quality of such DW images may be inferior to the quality of images obtained with a 1.5-T system; therefore, the ADC values of livers measured at b values of 0 and 500 to 600 second/mm2 may be significantly lower with a 3.0-T system.40 However, a different study demonstrated that the measurement of ADC at b values of 50, 400, and 800s/mm2 (as in the current study) provided ADC values comparable to those obtained with a 1.5-T system.37 Therefore, we believe that our results will be reproducible on 1.5-T systems.
In conclusion, we have demonstrated that HCCs with MVI have higher SI ratios on DW images obtained with b values of 50, 400, and 800 second/mm2 and lower ADC values in comparison with HCCs without MVI. A histological grade of 3 and an ADC value ≤ 1.11 × 10−3 mm2/second were independent predictors of MVI. Therefore, a higher SI ratio on DW images and a lower ADC value may be useful predictors of MVI during the preoperative evaluation of HCCs. If these findings are validated, the prediction of MVI with preoperative DWI might influence the selection criteria for hepatic resection and liver transplantation because the presence of MVI is related to poor resection and posttransplant outcomes.