Apparent diffusion coefficient of hepatocellular carcinoma on diffusion-weighted imaging: Histopathologic tumor grade versus arterial vascularity during dynamic magnetic resonance imaging

Objectives Apparent diffusion coefficient (ADC) has been suggested to reflect the tumor grades of hepatocellular carcinomas (HCCs); i.e., it can be used as a biomarker to predict the patients’ prognosis. To verify its feasibility as a biomarker, the present study sought to determine how the ADC values of HCC are affected by a tumor’s histopathologic grade and arterial vascularity. Materials and methods From 131 consecutive patients, 141 surgically resected HCCs (16 well-differentiated [wd-HCCs], 83 moderately-differentiated [md-HCCs], and 42 poorly-differentiated HCCs [pd-HCCs]) were subjected to a comparison of the tumors’ arterial vascularity (non-, slightly-, or markedly-hypervascular) determined on dynamic magnetic resonance imaging (MRI) and the ADC was measured retrospectively. Results The pd-HCCs (1.05±0.16 × 10−3 mm2/s) had a significantly lower ADC than md-HCCs (1.16±0.21 × 10−3 mm2/s; p = 0.010), but there was no significant difference compared to wd-HCCs (1.11±0.18 × 10−3 mm2/s; p = 0.968). The mean ADC was significantly higher in markedly hypervascular lesions (1.20±0.20 × 10−3 mm2/s) than in nonhypervascular lesions (0.95±0.14 × 10−3mm2/s; p<0.001) or slightly hypervascular lesions (1.04±0.15 × 10−3mm2/s; p<0.001). The ADC values and arterial vascularity were significantly correlated in wd-HCCs (p = 0.005) and md-HCCs (p<0.001). The mean ADC of pd-HCCs was significantly lower than those of other lesions, even in the markedly hypervascular lesion subgroup (p = 0.020). Conclusion Although pd-HCC constantly shows low ADCs regardless of arterial vascularities, ADCs cannot stably stratify histopathologic tumor grades due to the variable features of wd-HCCs; and the ADC should be used with caution as a tumor biomarker of HCC.


Introduction
Despite improvements in diagnosis and patient management, hepatocellular carcinoma (HCC) is the second leading cause of cancer-related deaths [1]. Poorly-differentiated (pd)-HCC have a poor prognosis because of its higher recurrence rate compared to well-differentiated (wd)-HCC and moderately-differentiated (md)-HCC [2][3][4]. The 5-year recurrence rate of pd-HCC is up to 75% [5] after surgical resection and 8-20% after liver transplantation [6]. Preoperative prediction of histopathologic differentiation for HCC is helpful for optimal treatment planning [7,8].
Diffusion-weighted imaging (DWI) with apparent diffusion coefficients (ADCs) obtained from DWI data has been introduced in abdominal imaging and is becoming a standard protocol of diagnostic magnetic resonance imaging (MRI) of the liver. This technique provides visualization of random Brownian motion of water molecules [9,10]. The use of DWI initially focused on the detection of tiny liver metastases by its synergistic effect with T2-weighted imaging and dynamic enhanced imaging [11][12][13][14][15][16]. It is also a promising technique for tissue characterization, based on the diffusional properties of water molecules through biologic tissues [17]. Some studies have demonstrated that DWI aids in distinguishing early HCCs from benign regenerative nodules in patients with chronic liver disease [18,19]. Moreover, DWI has been suggested to predict the histopathologic grade of malignant hepatic tumors because of an inverse correlation between ADC values and tumor grades [20][21][22][23][24][25][26][27]. However, in clinical practice, we have encountered exceptional cases of early lesions of hepatocarcinogenesis showing low ADCs. No consensus has been reached on this issue. The ADC inherently contains the perfusion fraction, which could be affected by tumor vascularity; hence, the contribution of tumor enhancement cannot be excluded in determining the ADC of HCCs. To verify its feasibility as a biomarker, the present study sought to determine how the ADC values of HCC are affected by a tumor's histopathologic grade and arterial vascularity.

Materials and methods
This retrospective study was approved by the Gangnam Severance Hospital institutional review board (IRB) for clinical studies, and the requirement for informed patient consent was waived.

Patients and clinical data collection
We searched the electronic medical records and radiological computer records of all patients surgically diagnosed with HCC at our institution between January 2011 and December 2015, and found 304 potential candidates. The inclusion criteria were as follows: underwent preoperative liver MRI (including dynamic enhanced imaging and DWI) within 2 months before surgery; location of tumor on MRI was the same as the surgically resected HCC in the pathologic report; the pathologic report described the histopathologic grade, based on the Edmonson and Steiner grading system; quality of dynamic enhanced imaging allowed evaluation of arterial vascularity; and definite localization of the lesion on DWI made it possible to draw regions of interest (ROIs). The exclusion criteria were as follows: lesion size was < 1 cm (to avoid the partial volume averaging effect); image was degraded because of cardiac motion or susceptibility artifact around the area of the lesions; evaluation of tumor's vascularity was difficult due to gross vascular invasion; or tumor was treated with transcatheter arterial chemoembolization or radiofrequency ablation before imaging. From 131 patients ranging in age from 38 to 81 years (mean age, 58.1 years, men:women, 104:27), 141 lesions were finally included (Fig 1). Most patients had underlying liver disease: chronic hepatitis B (n = 108), hepatitis C (n = 9), or hepatitis B and C with or without cirrhosis (n = 1) and alcoholic cirrhosis (n = 2), while the other 11 patients had no known underlying liver disease.

MRI protocol
We used a 1.5-T MRI system (Magnetom Avanto, Siemens, Erlangen, Germany) fitted with high-performance gradients (maximum amplitude 45 mT/m) and a six-element phased-array surface coil. After the non-contrast fast T2-and T1-weighted MRI with use of half-Fourier single-shot turbo-spin-echo and double-echo chemical shift gradient echo sequences, dynamic MRI was performed. A fat-suppressed three-dimensional (3D) gradient echo sequence (i.e., volumetric interpolated breath-hold examination [VIBE]; Siemens, Erlangen, Germany) was obtained before injecting the intravenous contrast media of gadoxetic acid (Primovist, Bayer Schering; 0.025 mmol/kg). After a 1-mL test-bolus injection of gadoxetic acid to determine the timing of the earliest phase, the contrast media was injected through a power injector at a speed of 1 mL/s, followed by a 20-mL physiological saline flush at the same rate. Dynamic Apparent diffusion coefficient of hepatocellular carcinoma: Tumor grade vs arterial vascularity imaging, including the early arterial phase, late arterial phase, and portal venous phase, was performed at 34-s intervals (20 s for image acquisition with breath-holding and 14 s for rebreathing). Corresponding subtraction arterial-dominant phase images were acquired with coregistration software (Inline Liver registration; Siemens Medical Solutions, Erlangen, Germany), which performed image-by-image subtraction of the precontrast images from the coregistered postcontrast images.
Respiratory-triggered DWI with single-shot echo planar imaging was then acquired with motion-probing gradients in three directions. Respiration was monitored using the prospective acquisition correction technique, which periodically evaluates diaphragmatic position by navigator echoes. The scanning parameters for DWI were two b factors, 50 s/mm 2 and 800 s/ mm 2 ; TR, 3900 ms; TE, 75 ms; matrix size, 156×192; average, 6; 54-60 slices (27-30 for each b factor); slice thickness, 6 mm; and interslice gap, 20%. The MRI system automatically calculated the ADC values for each DWI sequence and generated the corresponding ADC maps. The delayed hepatobiliary phase with the VIBE sequence was obtained 20 min after contrast injection.

Data analysis
The study coordinator (an abdominal radiologist with 20 years of experience) preliminarily reviewed all images, pathologic reports, and medical records. Each HCC was marked with an arrow on the gadoxetic acid-enhanced hepatobiliary phase images. For the precontrast T1-weighted images and arterial-dominant (early or late arterial) phase images combined with the corresponding subtraction images, two radiologists (a radiologist with 10 years of experience in abdominal MRI and a third-year resident) were blinded to the histologic grade of the lesions. They independently determined the arterial vascularity of each tumor using a threepoint scoring system: a score of 1 (i.e., nonhypervascular) indicated a hypointense lesion on arterial-dominant phase images, regardless of the findings of the subtraction images, and a hypointense or isointense lesion on the corresponding subtraction images for hyperintense lesions on precontrast T1-weighted images; a score of 2 (i.e., slightly hypervascular) indicated an isointense or slightly hyperintense lesion on arterial-dominant phase images that was a hyperintense lesion on the corresponding subtraction images; and a score of 3 (i.e., markedly hypervascular) indicated a lesion that was vividly hyperintense on arterial-dominant phase images and the corresponding subtraction images. When the two radiologists disagreed on the degree of tumor enhancement, two reviewers together reviewed the images and re-evaluated the vascularity for future analysis.
The ADC of each tumor was measured independently by the same two radiologists by using an oval or polygonal ROI on the ADC map. The largest ROI was placed on the solid portion where the previously assessed vascularity was determined on dynamic imaging. For nonhypervascular lesions, the ROI was placed on the non-necrotic solid portion by referring to other sequence images, including the precontrast T1-and T2-weighted images and DWI. The tumor's margin was excluded to minimize the partial volume average effect. When the lesion was small and indistinguishable from surrounding hepatic parenchyma on the ADC map, the corresponding DWI images (b value, 50 s/mm 2 ) were simultaneously displayed on picture archiving and communication system monitors. Two horizontal and perpendicular lines passing the center of the lesion were drawn from the left border and the upper border of the image to determine the x and y coordinates of the lesion on the DWI image. Two identical lines were drawn on the corresponding ADC map. The ROI was placed around the manually synchronized center of the lesion to measure the ADC values. The average values obtained by two radiologists were used for further analyses.
The pathologists reported the major and worst Edmonson and Steiner grades of HCCs for all lesions in the patients' medical records. In the present study, the major grade was regarded to represent the imaging characteristics of each lesion. All lesions were divided into three histopathologic groups for future analysis: 16 wd-HCCs for Edmonson and Steiner grade 1 lesions; 83 md-HCCs for grade 2 lesions; 42 pd-HCCs comprising grade 3 (35 lesions) and grade 4 (7 lesions).

Statistical analysis
Interobserver agreement regarding the degree of arterial phase contrast-enhancement was assessed using linear-weighted Cohen kappa tests. A kappa value of 0.00-0.20 indicated slight agreement; 0.21-0.40, fair agreement; 0.41-0.60, moderate agreement; 0.61-0.80, good agreement; and 0.81-1.00, excellent agreement. The Bland-Altman test was performed for the reproducibility of ADC measurement between the two observers. To ensure that the number of HCC in each group was sufficient to draw conclusion, ANOVA post-hoc power calculation was carried out using Power Analysis and Sample Size 12 for Windows software package (NCSS Inc, LLC, Kaysville, UT, USA). After using the Kolmogorov-Smirnov test to define the data distribution pattern of the measured ADC values, the relationships between the mean ADCs and two factors (i.e., the histopathologic grades and the arterial vascularity) were stratified using analysis of variance (ANOVA) tests. To verify the relationship of ADCs in each subgroup of histopathologic tumor grade for arterial vascularity and vice versa, ANOVA with Bonferroni corrected post-hoc analysis was performed. All statistical analyses were performed using SPSS Statistics version 23 (IBM Corp., Armonk, NY). Statistical significance was set at P < 0.05.

Interobserver agreement and data distribution
The Cohen kappa test revealed good interobserver agreement in the degree of arterial phase contrast-enhancement (mean kappa value = 0.770). The Bland-Altman plot (Fig 2) showed no fixed or proportional bias in the ADC assessment. In ANOVA post-hoc power calculation, the power was 82.59% between wd-HCCs, md-HCCs and pd-HCCs. Substantial reproducibility of ADC measurement between the two observers was proven. The mean absolute difference in the ADC measurement between the two observers was 0.004 × 10 −3 mm 2 /s (limit of agreement, 0.20-0.66). There was no null distribution (P > 0.05). The overall ADC data showed sufficient normality to fit the subsequent statistical analyses.

Subgroup analysis of tumor grades with arterial vascularities
For the relationship of ADCs in each histopathological tumor grade subgroup with arterial vascularity, score 3 lesions showed higher ADCs than score 1 lesions (P = 0.023, P < 0.001) or score 2 lesions (P = 0.021, P = 0.001) in the subgroups of wd-HCC and md-HCC, respectively. Only the score 3 lesions showed a statistically significant difference of mean ADCs in histopathologic tumor grade (md-HCC vs. pd-HCC, P = 0.017) (Tables 1 and 2).

Discussion
Similar to earlier reports on genitourinary malignancy (e.g., bladder and cervix) using the ADC of DWI for tumor grading [28,29], the potential use of ADC for depicting the histopathologic grade of HCC has also been suggested [20][21][22][23][24][25][26][27]. In the present study, however, there was Apparent diffusion coefficient of hepatocellular carcinoma: Tumor grade vs arterial vascularity no significant stratification of mean ADCs according to the tumor grades. The finding of comparable or lower ADCs of wd-HCCs, compared to those of md-HCCs, coincided with the prior results of Nasu et al [30], which showed that the histopathologic grade of HCC was not correlated with the ADC. As speculated [30], the cellular atypia of different nucleus-cytoplasm ratios, which primarily determined the histopathologic tumor grade, would be trivial in revealing gross differences in ADC on current DWI [31,32].
Previous research on the correlation between arterial blood supply and grades of malignant hepatic nodules demonstrated that arterial blood supply increases in the early stage of HCC development and then decreases in the late stage [33]. Therefore, the ADC of HCC in DWI may reflect the degree of tumor vascularity rather than the histopathologic tumor grade, which supports the finding that md-HCCs showed higher ADC values than wd-HCCs and pd-HCCs because ADC is not only affected by molecular diffusion but also by microcapillary perfusion, especially by use of small b factors [34,35]. Even when we used b = 50 instead of b = 0 for the ADC calculation to reduce the perfusion effect during DWI, the perfusion effect was still large enough to have a considerable influence on the ADCs in the present study.
characteristics of wd-HCCs (Fig 4). In clinical practice, we have often experienced wd-HCCs showing either the typical appearance of classical HCC with hypervascularity and T2-weighted hyperintensity or the nonhypervascularity with T1-weighted iso-/hyper-intensity and T2-weighted iso-/hypo-intensity mimicking regenerating or dysplastic nodules. The fact that pathologists could use subjective criteria on a case-by-case basis in the differential diagnosis of borderline malignant cirrhotic nodules, including wd-HCC and dysplastic nodules, and in tumor grading may have also contributed to this contradiction [36]. Locating the ROI in the tumor and defining the representative tumor grades for rather heterogeneous tumor components may be debatable. Some investigators used minimal ADCs among the values measured on multiple ROIs in different locations of each tumor [22]. However, there was no significant difference between the minimal ADCs and average ADCs of the solid component as a representative value in a comparative analysis with histopathologic tumor grades [27]. If a necrotic or hemorrhagic tumor portion could be excluded from the ROI measurement, inadequate measurement or sampling error would be of no remarkable concern. In the present study, the subgroup analysis of histopathologic tumor grades in each arterial vascularity group and vice versa revealed that ADC significantly increased with the arterial vascularity in the wd-and md-HCCs. However, the incremental degree was not sufficiently large to show statistical significance in the pd-HCCs (Fig 5). We did not have ADC data using b = 0, which could be more sensitive for perfusion fraction; these findings indicated that the different degrees of arterial vascularity were insufficient to grossly influence the ADC values in pd-HCCs. The structural atypia of high cellular density with distortion of extracellular stroma restricting the diffusion of water molecules in the intercellular spaces could be more influential than the degree of arterial vascularity in the pd-HCCs. Similar to prior reports [20][21][22][23][24][25][26][27], the consistently low ADC value in pd-HCCs may arise from the synergistic effect of low vascular perfusion and restricted water diffusion.
Recently, there have been several reports on the use of intravoxel incoherent motion (IVIM) imaging to determine the correlation of diffusion and perfusion parameters with histologic grades of HCC using a biexponential algorithm [37][38][39]. Despite prior studies suggesting the usefulness of IVIM for the prediction of histologic grades of HCC, there were inconsistencies in the findings of these studies [37][38][39]. A standardized protocol and a large study are needed to thoroughly test the suitability of IVIM for this purpose.
Our study has several limitations. First, arterial vascularity was not histologically assessed, and there were no real-time data of tumor perfusion. In a previous study involving IVIM, the Apparent diffusion coefficient of hepatocellular carcinoma: Tumor grade vs arterial vascularity subjectively determined arterial vascularity-on a three-grade scale similar to the one used in the present study-was well correlated with the perfusion fraction determined on IVIM [37]. Based on the results of another study showing the relationship of the arterial enhancement during dynamic imaging with tumor vascularity [40], we subjectively determined the arterial vascularity in the present study. Despite a combined review of dynamic images with corresponding subtraction images and good interobserver agreement, more standardized quantitative data of continuous variables for perfusion are needed to reinforce the results of the present study. Second, the major histopathologic tumor grade was used for the analysis in the present study. However, the patients' prognosis depended on the worst grade in the HCCs; in the preliminary analysis, however, the worst tumor grades did not show any statistically significant results. This finding may indicate limited value of DWI as a biomarker for predicting the tumor prognosis of HCCs.
In conclusion, the ADC of HCCs could not stratify the histopathologic tumor grades, probably because of the variable ADC and arterial vascularity of wd-HCCs and the insufficient difference of diffusivity between the wd-and md-HCCs. The degree of arterial phase enhancement might be rather well correlated with the ADC values for wd-HCCs and md-HCCs while pd-HCCs show constantly low ADCs regardless of arterial vascularities. Based on the results of the present study, hypervascular HCCs with low ADC could be interpreted as pd-HCCs with poor prognosis, while it remains difficult to differentiate pd-HCCs from wd-HCCs for Apparent diffusion coefficient of hepatocellular carcinoma: Tumor grade vs arterial vascularity nonhypervascular lesions using only the ADC values. Due to the limited ability to stratify the histopathologic tumor grades, the ADC should be used with caution as a tumor biomarker in determining the prognosis of HCCs.