Diagnostic Value of Imaging Methods in the Histological Four Grading of Hepatocellular Carcinoma

We attempted to establish an ultrasound (US) imaging-diagnostic system for histopathological grades of differentiation of hepatocellular carcinoma (HCC). We conducted a retrospective study of histopathologically confirmed 200 HCCs, classified as early (45 lesions), well- (31 lesions), moderately (68 lesions) or poorly differentiated (diff.) (56 lesions) HCCs. We performed grayscale US to estimate the presence/absence of halo and mosaic signs, Sonazoid contrast-enhanced US (CEUS) to determine vascularity (hypo/iso/hyper) of lesion in arterial and portal phase (PP), and echogenicity of lesion in post-vascular phase (PVP). All findings were of significance for the diagnosis of some (but not all) histological grades (p < 0.001–0.05). Combined findings with a relatively high diagnostic efficacy for early, poorly and moderately diff. HCC were a combination of absence of halo sign and isoechogenicity in PVP of CEUS (accuracy: 93.0%, AUC: 0.908), hypovascularity in PP (accuracy: 78.0%, area under the curve (AUC): 0.750), and a combination of isovascularity in PP and hypoechogenicity in PVP (accuracy: 75.0%, AUC: 0.739), respectively. On the other hand, neither any individual finding nor any combination of findings yielded an AUC of over 0.657 for the diagnosis of well-diff. HCC. Our study provides encouraging data on Sonazoid CEUS in the histological differential diagnosis of HCC, especially in early HCC, and the effectiveness of this imaging method should be further proved by prospective, large sample, multicenter studies.


Introduction
A histological classification method for hepatocellular carcinoma (HCC) can provide valuable information for determining the prognosis and selecting appropriate treatments, and perhaps even

Patient Enrollment
A total of 200 consecutive liver biopsy specimens from 157 patients seen between January 2014 and August 2018 were analyzed in this study. All the lesions were newly discovered and as yet untreated, with histopathological confirmation of both the diagnosis of HCC and the histopathological grade of HCC. Clinical information (gender, age, etiology of cirrhosis, Child-Pugh grade), imaging data, and histology reports of the patients were collected by a retrospective review of the electronic medical records, radiology database, and pathology records of our hospital, respectively. Our retrospective study design was approved by the institutional review board of the Medical Center of Yokohama City University (Number B180200054) and in compliance with the principles of the Declaration of Helsinki; informed consent for the study was obtained from each of the participating patients. The criteria for exclusion from the present study were as follows: 1.
Child-Pugh grade C liver cirrhosis.

2.
Liver CEUS not performed within one month prior to the biopsy.

3.
No reliable or conclusive pathological diagnosis could be obtained because of insufficient biopsy specimens from the lesion, or a "no-hit" result for the target lesion during a percutaneous biopsy.

4.
Systemic chemotherapy or targeted treatment administered prior to the CEUS, which could potentially influence the findings on the CEUS images.

Grayscale US and Sonazoid CEUS Examination
Grayscale US and CEUS of the liver were performed at least once, within a month prior to the percutaneous US-guided liver biopsy. The equipment used was a LOGIQ E9 US system (GE Healthcare, Milwaukee, WI, USA) equipped with a native tissue harmonic grayscale imaging and CEUS function. The probes used were a convex probe with a frequency of 1-6 MHz and a micro-convex probe with a frequency of 2-5 MHz.
At first, grayscale US was performed. The size and position of the target hepatic lesion, the echogenicity of the lesion, the presence/absence of the internal mosaic sign and peripheral halo sign were observed and recorded. The diagnosis of "high", "iso", or "hypo" echogenicity was made on the basis of a more than 50% area of the lesion showing "high", "iso", or "hypo" echogenicity, respectively, as compared to the surrounding liver parenchyma.
Then, Sonazoid CEUS was performed. A 0.2-mL dose of Sonazoid (Daiichi Sankyo, Tokyo, Japan) was injected into an antecubital vein at 0.2 mL/s via a 24-gauge cannula, followed by 2 mL of 5% glucose. CEUS images were acquired and evaluated during three contrast phases, namely, the arterial phase (AP) (10-50 s after initiation of injection), the portal phase (PP) (80-120 s after initiation of injection), and the post-vascular phase (PVP, also called the Kupffer phase) (10 min after initiation of injection). The vascularity of the lesions, as compared to that of the surrounding liver parenchyma, was classified as hyper-, iso-, or hypovascularity, when more than a 50% area of the lesion showed the above patterns, respectively. In general, the low mechanical index (MI) mode (0.21-0.28) (Figure 1d, Figure 2b,c, Figure 3c,d and Figure 4d) was used for the imaging, as it allows the entire liver to be observed in real time. Sometimes, we manually switched to low MI (0.18-0.28) harmonic imaging for the CEUS, which allows a more detailed evaluation of the tumor vessels and tumor staining, due to the high frame rate [22,23] (Figure 1b In cases with hyperechoic lesions whose observation may be influenced by the background grayscale mode, we used high-MI (0.7-1.2) contrast imaging for more accurate evaluation of the presence/absence of a defective area during the PVP [23] (Figures 1e, 2d and 3e).       The US and CEUS examinations of all patients were performed together by two doctors who have been working in the field of liver US diagnosis for seven and 20 years, respectively. In view of the relative subjectivity of interpretation of the mosaic sign and halo sign in the US images, the presence/absence of these signs in each case was discussed and determined by two doctors. The CEUS images were also reviewed independently by the two doctors, a sonologist and a hepatologist, who had worked in the field of US diagnosis and liver disease for seven and 20 years, respectively. They were blinded to the final diagnosis, and the clinical and other radiological information about the patients.

Gd-EOB-DTPA MRI
The majority of the enrolled patients (with 157/200 lesions, 78.5%) underwent Gd-EOB-DTPA MRI of the upper abdomen, within a month prior to the liver biopsy. A few patients did not undergo Gd-EOB-DTPA MRI, either because it was contraindicated (metallic implants in the body) and/or it was not necessary (HCC was clearly diagnosed by enhanced CT). The MRI was performed in a 1.5or 3-T MR system (Signa HDx; GE Healthcare). Each patient received a rapid intravenous bolus injection of 0.25 mmol/kg body weight of gadoxetic acid (Gd-EOB-DTPA, Primovist ® , Bayer Schering Pharma, Berlin, Germany) at the rate of 1 mL/s, followed by a 20-mL saline flush. Dynamic multiphasic (arterial-, portal-, and transitional-phase) imaging was performed using a fat-suppressed 3D T1WI gradient echogenicity sequence. The hepatobiliary phase (HBP) images were obtained 20 min after injection of contrast material. The signal intensity of each of the hepatic lesions, as compared to that of the surrounding liver parenchyma, was evaluated and the lesion was recorded as being a hypo-, isoor hyperintensity when all or more than a 50% area of the lesion was visualized as a hypo-, iso-or hyperintensity, respectively.

Diagnostic Criteria
After confirming the location of the HCC lesions by grayscale US/CEUS and Gd-EOB-DTPA MRI, a fine-needle aspiration biopsy was performed using a 21-gauge fine biopsy needle (SONOPSY; Hakko, Tokyo, Japan), under US/CEUS guidance. At least two samples from each lesion and one sample from the adjacent liver tissue (negative control) were obtained. All the tissue specimens were formalin-fixed, paraffin-embedded, and consecutively sectioned to the same thickness; the sections were stained first with HE, and then silver and VB stains, followed by immunostaining for CD34 and CK7.
staining showed no stromal invasion of cancer cells. (i) Cytokeratin 7 staining showed no expression of the marker in the tumor areas, that was an absence of ductular reaction. (j) Diffuse expression of CD34 in the peripheral areas suggested increased neovascularization resulting from sinusoidal capillarization and a formation of sinusoidal vascular endothelium in HCC. This lesion was histopathologically diagnosed as a moderately diff. HCC. The arrowheads seen in (a)-(e) indicated the margin of the lesion. Another lesion was also seen in (a), (b), (d), (e) (arrows).  The HCCs were classified histologically according to the International Working Party criteria [12] and ICGHN [13], into early, well-diff., moderately diff., or poorly diff. HCC. We took well-, moderately and poorly diff. HCCs as a whole to be advanced HCCs. The features examined were the degree of cellular atypia, structural atypia, pattern of the reticular fibers, degree of neovascularization, degree of stromal invasion, and the ductular reaction ( Table 1). The cell density, the nuclear size and morphology, the nuclear-cytoplasmic ratio, and alterations of the trabecular structure were evaluated in the sections stained with HE. (Figures 1g, 2f, 3f and 4i). A decrease in the reticular fibers was evaluated using silver staining (Figures 1h, 2g, 3g and 4j). The ICGHN arrived at the following consensus regarding the histological diagnostic criteria for early HCC [13]: it stated that the presence of tumor cell invasion into the intratumoral stroma (portal tract) [24] should be recognized as the most important histopathologic finding for the diagnosis of early HCC. The diagnosis of intratumoral stromal invasion requires VB staining [25] (Figure 1i). The ductular reaction, which refers to the staining of bile ducts in the periportal tract resulting from the change of bile metabolism in cases of HCC, was evaluated by immunostaining for CK7 [26] (Figure 1j). HCC sinusoids acquire the characteristics of capillary and precapillary blood vessels during de-differentiation from early to advanced HCC, and the tumor begins to reveal hypervascularity on angiography [27]. Neovascularization was examined by immunostaining for CD34, which is one of the most commonly used endothelial cell markers [27].
Immunohistochemical staining for CD34 (monoclone: QBEnd/10, mouse anti-Human; supplier: NOVO, Corp. Carpinteria, CA, USA; dilution: 1:100) and CK7 (monoclone: OV-TLJ 12/30, mouse anti-Human; supplier: DAKO, Corp., Carpinteria, CA, USA; dilution: 1:800) was performed using the LSAB kit, in accordance with the manufacturer's instructions. It should be noted that CK7 can stain the following: (1) the cytoplasm of a few activated cancer cells (progenitor cells and intermediate hepatocytes) in some aggressive HCCs [28,29]; and (2) bile ducts around the portal tract (periportal tract) within the HCC lesion [26,29]. As bile-producing metabolic pathways in the malignant cells are abnormal in HCC, the expression of CK7 in the periportal bile ducts within the HCC lesion is decreased (early HCC) or absent (early HCC and advanced HCC) in HCC as compared to the findings in the adjacent, non-tumor parts of the liver. In line with the purpose of the study of histological grading of HCC, the observation of CK7 staining in the current study was confined to the bile ducts within the HCC lesion. CD34 staining was considered positive when positive staining was noted in the cytoplasm/membranes of vascular endothelial cells other than cancer cells. The CD34 staining pattern was classified as negative, focal positive or diffuse positive. "Negative" was defined as positive staining of only the blood vessels and bile ducts in the portal tracts and/or rare positive staining of the sinusoidal spaces (<10%) near the portal tracts. This pattern was observed in both the non-tumor areas and cases of early HCC. "Focal" was defined as positive staining of a small proportions (10-50%) of the sinusoidal spaces (Figure 1k). "Diffuse" was defined as positive staining of the majority of sinusoidal spaces (>50%) throughout the lesion area (Figure 2j, Figure 3j, and Figure 4m).  2 Here, CD34-positive expression refers to positive staining of the sinusoidal capillaries. Focal and diffuse expressions of CD34 were defined as positive staining of only a proportion of the sinusoidal spaces (less than 50%) and staining of the majority of sinusoidal spaces (>50%) throughout the lesion, respectively; 3 Here, VB positive staining means staining of the portal tracts in the cancer lesion; thus, stromal invasion was defined as the presence of cancer cells within the portal tracts. If portal tract invasion was recognized within the HCC lesion, the grade of cancer cell atypia was irrelevant; 4 Here, as compared to the normal number of bile ducts around the portal tract (periportal bile ducts) in the non-tumor area of the liver, few or no periportal bile ducts (as determined by CK 7 immunohistochemistry) in the cancer lesion was classified as "decreased" or "absent" ductular reaction.
The final histopathological diagnosis was made by the consensus of two expert pathologists with over 20 years of experience in liver pathology. In particular, one of the expert pathologists was a member of both the International Working Party and the ICGHN. Both the pathologists were blinded to the clinical, laboratorial, and imaging findings of the patients. Descriptive rather than numeric designations were used. Especially, when varying grades of differentiation were observed in different areas within a single tumor, we determined the diff. grade according to the worst diff. grade observed within the tumor [7].

Statistical Analysis
All the indicators examined in this research (baseline characteristics, imaging patterns, and final pathological diagnoses) were statistically analyzed.
Age as a continuous variable (with normal distribution) was presented as the means ± standard deviation (SD), and the mean ages were compared among the four histological groups by one-way analysis of variance (ANOVA). Tumor size as a continuous variable (not with normal distribution, but with homogeneity of variance) was presented as the median values (interquartile range) and compared among the four histological groups by a non-parametric Kruskal-Wallis test. The remaining baseline data and imaging indicators were tested among the four histological groups by the chi-square test, chi square Linear-by-Linear association analysis, and Fisher's exact test.
Two-tailed Fisher's exact test and chi-square test were used for comparison of the frequencies of all the categorical variables to determine inter-group differences of the histological grade, where appropriate. The area under the curve (AUC) of the receiver operating characteristic curve was used to determine the diagnostic usefulness of indicators for each differentiation grade of HCC. The level of significance was set at p < 0.05. The cutoff values of continuous data (size of the lesion) were obtained from the receiver operating characteristic (ROC) curve analysis. The size value yielding the maximum sum of sensitivity and specificity is determined as the cutoff value. The above statistical analyses were performed using SPSS version 26.0 for Windows (SPSS Inc., Chicago, IL, USA). Overall missing or implausible data were excluded from the statistical analysis and extreme laboratory values were counted as missing. The sensitivity, specificity, and accuracy of each of the indicators were calculated manually. Table 2 shows the baseline characteristics of the study population with early, well-diff., moderately diff., and poorly diff. HCC. Of the total number of patients with HCC patients, 29, 111, and 60 patients had underlying HBV, HCV and other diseases (such as non-alcoholic steatohepatitis and primary biliary cirrhosis), respectively. There were no significant differences in the general characteristics (age, sex, etiology of HCC, Child-Pugh classification, location of the lesions) among the groups. The mean diameters (median (interquartile range)) of the early, well-diff., moderately diff. and poorly diff. HCCs were 14.00 (12.00-16.00), 17.00 (15.00-20.00), 17.50 (15.00-23.50), and 20.00 (12.00-26.50) mm, respectively. The size of lesion was defined by the maximum diameter measured in the US images (with reference to Gd-EOB-DTPA MRI and CEUS). Eight lesions showed a significant deviation of size (>100 mm in diameter) relative to that of the rest of the lesions and were beyond the measurement range of the US display screen. Considering the significant measurement error, they were regarded as invalid values and excluded from the statistical analysis of the sizes. The tumors differed significantly in size among the histological groups (p < 0.001, Table 3).  3 Age showed normal distribution in each group and was tested by single factor analysis of variance (ANOVA). Tumor sizes were compared among different histological grades using the non-parametric Kruskal-Wallis test. The rest of the data were tested by the chi-square test or chi square Linear-by-Linear association analysis.

Comparison of the Grayscale US Patterns, and the CEUS and Gd-EOB-DTPA MRI Enhancement Patterns among the Four Histological Grades
When each of the imaging findings was analyzed independently (Table 3, p-value), the lesion size, presence/absence of the halo sign and mosaic sign in the grayscale US images, lesion intensity in each of the three phases of CEUS and in the HBP of Gd-EOB-DTPA MRI were significantly different in some histological grades. Specifically, the lesion size, presence/absence of the halo sign, and the lesion intensity in the AP and PVP allowed clear differentiation of early HCC from the other histological grades. Differences in the vascularity in the PP of CEUS were observed between poorly diff. HCC and all other histological grades of HCC. Between well-diff. and moderately diff. HCC, there were statistically significant differences in the positivity rate for the halo and mosaic signs, and findings in the PVP of CEUS. However, the findings on Gd-EOB-DTPA MRI allowed only differentiation between early HCC and poorly diff. HCC (p = 0.027). The echogenicity of the lesions on grayscale US was revealed to be of no significant value in the differential diagnosis of the histological grades of HCC (Figures 1a, 2a, 3a  and 4a).

Analysis of the Diagnostic Efficacy of Each of the Findings
Some of the findings in our study showed perfect-to-good efficacy for the diagnosis of early HCC (Table 4) and poorly diff. HCC ( Table 4). The diagnostic efficacy for well-diff. and moderately diff. HCC (Table 4) was not very high (details of the calculation are not listed in this article). The best method to diagnose moderately diff. HCC was to determine the lesion vascularity in the PVP of CEUS, which was associated with an accuracy of 62.5% and an AUC of 0.697. Unfortunately, none of the imaging findings, either alone or in combination, showed an AUC of more than 0.657 or of any value for the diagnosis of well-diff. HCC (Table 4, data not completely shown).  The isoechoic pattern in PVP of CEUS showed the best efficacy for the diagnosis of early HCC (Figure 1d,e), with a sensitivity and specificity of 91.1% and 91.6%, respectively. The combination of isoechogenicity in PVP of CEUS and absence of the halo sign on grayscale US yielded the highest accuracy (96.1%) and a relatively high AUC (0.930) for the diagnosis of early HCC as compared to any imaging findings for any other histological grades.
A hypovascular pattern in PP of CEUS was the most valuable diagnostic indicator for poorly diff. HCC (Figure 4c). The sensitivity, specificity, and accuracy of the hypovascular pattern in PP of CEUS for the diagnosis of poorly diff. HCC were 71.7%, 80.3%, and 78.0%, respectively. However, an addition of the US findings in the AP of CEUS or PVP of CEUS or any other US findings for the diagnosis of poorly diff. HCC did not improve the efficacy, especially the accuracy and the AUC. Furthermore, as compared to other findings, the findings of the hypovascular pattern in PP of CEUS yielded the highest accuracy (88.7%) and AUC (0.780) for the diagnosis of poorly diff. HCC.
For the diagnosis of moderately diff. HCC (Figure 3b-e), the combination of isovascular patterns in the PP and hypoechogenicity in PVP of CEUS was associated with acceptable diagnostic efficacy, with an AUC of 0.739 and an accuracy of 75.0%. As compared to other indicators for the diagnosis of well-diff. HCC (Figure 2a-c), the combination of hypervascularity in AP of CEUS and the isovascular pattern in PP of CEUS showed a relatively high diagnostic efficacy for well-diff. HCC, with an accuracy and an AUC of 62.0% and 0.657, respectively.
The cutoff value of the lesion size obtained from the ROC curve analysis was 17.5 mm. Using this cutoff value, the lesions were divided into two groups: lesions measuring ≥18 mm and lesions measuring less than 18 mm in diameter. The lesion diameter differed significantly only between early HCC and other histological groups (Table 3, p < 0.001). Therefore, we conducted a subgroup analysis to determine the diagnostic value of each of the imaging findings for early HCC according to the size; the subgroup analysis showed that all the imaging findings that were found to be useful for the diagnosis of early HCC were still significant after adjustment for the lesion size (p < 0.05, data not shown). In particular, the AUC and accuracy of the most valuable diagnostic indicators (a combination of the halo sign and isovascularity in the PVP) were still high. The AUC for the subgroup with the smaller tumor size was 0.900, while that for the subgroup with the larger tumor size was 0.897. The diagnostic accuracy for the subgroup with the smaller tumor size was 90.0%, while that for the subgroup with the larger tumor size was 95.2% (Table 4).

Discussion
In our study, we found that using some imaging indicators alone or in combination yielded a high diagnostic efficacy for early HCC, as well as acceptable efficacy for poorly diff. and moderately diff. HCC. The diagnostic efficacy of the imaging indicators used alone or in combination proved to be much lower for well-diff. HCC as compared to that for the other three histological grades. Taken together, use of indicators derived from Sonazoid CEUS, such as the pattern of vascularity in the PP and echogenicity in the PVP, may be promising for diagnosis of the histological differentiation grade of HCC.
Grayscale US plays an important role in the diagnosis of HCC. Some of the latest clinical guidelines recommend that patients with hepatitis B, regardless of whether they have liver cirrhosis or not, must undergo regular US screening every six months [17,18]. A peripheral hypoechogenic "halo" of the lesion has been explained as being the radiological equivalent of a thin fibrous capsule in cases of advanced HCC [30]. It is believed to reflect the tendency towards expansive growth as large lesions begin to infiltrate the surrounding parenchyma as well as adjacent vessels and its branches [31]. The "mosaic" sign is positive in large tumors with a heterogeneous imaging appearance caused by multiple histological components (viable tumor with fat, necrosis and/or hemorrhagic components) [32], considered to be typical of advanced rather than small, early HCC. According to our study, both the "halo" and "mosaic" signs were useful to some extent for differential diagnosis of the histological differentiation grade of HCC. In particular, an absence of the "halo" sign was of significant usefulness for the diagnosis of early HCC, with a sensitivity as high as 95.65%. From the proportion of cases in the four histological groups showing positive "halo" (early: 4.4%; well-diff.: 41.9%; moderately diff.: 67.6%; poorly diff.: 57.1%) and "mosaic" (early: 2.2%; well-diff.: 16.2%; moderately diff.: 35.3%; poorly diff.: 35.7%) signs, the positive rate for "halo" signs revealed a sharply increasing trend with progression from early to moderately diff. HCC, while that for the "mosaic" sign seemed to show no such significant trend. Thus, our data suggested that the "halo" sign is of greater diagnostic value than the "mosaic" sign. On the other hand, our results showed that the echogenicity of a lesion was of no value for differentiating among the histological differentiation grades of HCC.
Hemodynamic changes taking place during hepatocarcinogenesis can be identified by CEUS. In general, the intranodular normal supply of hepatic arteries and portal veins decrease, while unpaired abnormal arteries and intranodular hepatic sinusoidal capillaries gradually increase in density with a worsening of the histological differentiation grade of HCC [33]. Almost all advanced HCCs are supplied entirely by the hepatic arterial system via abnormal unpaired arteries [34]. In our study, the findings in all three phases of Sonazoid CEUS were found to be of value in the differential diagnosis of the histological differentiation grade of HCC. Furthermore, in regard to specific analysis of correlation of the findings in each CEUS phase in relation to each grade, the efficacies were quite different. In regard to the findings in the AP, it has been reported that early and advanced HCCs classified on the basis of the histological grade usually show a hypovascular and hypervascular appearance, respectively [10,35]. In the PP and PVP, the washout time became shorter as the histological differentiation grade became poorer. Well-diff. HCC showed a more-delayed or no washout as compared to moderately and poorly diff. HCC [36][37][38].
The findings of our research are consistent with those of previous studies, but offer a more detailed differential diagnosis. From this study, we concluded that the most characteristic CEUS finding of early HCC is the isoechogenicity of the lesion in the PVP, suggestive of the absence of washout, in cases of early HCC. In contrast, washout began earliest (that is, in the PP rather than in the PVP) in cases of poorly diff. HCC, which can well explain the finding in our study that hypovascularity in the PP showed the best diagnostic efficacy for poorly diff. HCC. Notably, we found that the imaging findings in the PVP allow moderately diff. HCC to be differentiated from well-diff. HCC, whereas there were no significant differences in the findings in the AP or PP of CEUS between these two grades of differentiation. We also found that the findings of isovascularity in the PP and hypoechogenicity in the PVP can be of value in the diagnosis of moderately diff. HCC. Our results may suggest that the observer should pay more attention to the later phases of CEUS (PP and PVP) rather than to the earlier phase (AP). The strong ability of findings in the PVP of Sonazoid CEUS to allow different histological grades of HCC to be differentiated might be explained by the mechanism of progression of HCC: as HCC progresses, the Kupffer cells may become fewer or even disappear, resulting in an area clear of contrast material or a perfusion defect in Kupffer imaging [38]. This was further confirmed by the study of Imai, who used a kind of enhanced MRI to accurately determine the Kupffer cell numbers to predict the histological differentiation grade of HCC [39]. Nevertheless, we also found exceptions: for example, the presence of thrombosis in the portal vein or hepatic veins may contribute to vein flow reversal, exerting an influence on the hepatic circulatory dynamics and the diagnostic ability of CEUS [36]. The aforementioned explanations require to be verified through thorough and reliable data collection.
Gd-EOB-DTPA has extracellular properties, which allow the blood supply of a lesion to be well-visualized. The "washout" of contrast materials from HCC lesions, and the different degrees of "washout" are supposed to be attributable to a gradual decrease of the portal blood supply with progression of the histological grade of HCC [40]. Moreover, Gd-EOB-DTPA is well-recognized for its hepatocyte-selective properties, which allow the functions of hepatocytes to be evaluated. During hepatocarcinogenesis, the dysfunctional hepatocytes show impaired uptake of contrast agents, thus allowing for visualization of hypointensity of HCC lesions as compared to the surrounding normal functional hepatocytes. Theoretically, different histological differentiation grades of HCC could be diagnosed based on different degrees of decrease in the blood supply and dysfunction of hepatocytes. A previous study reported a correlation between the relative enhancement ratio (the signal intensity of the tumor relative to the adjacent underlying liver) in the HBP of Gd-EOB-DTPA MRI and the histological differentiation grade of HCC [20]. Some studies have successfully combined findings in the HBP in Gd-EOB-DTPA MRI with some other parameters (quantitative analysis based on T1 mapping, hyperintensity on high-b-value diffusion weighted imaging, hypervascularity in the AP of Sonazoid CEUS) for accurate diagnosis or differential diagnosis of different histological differentiation grades of HCC [41][42][43]. These studies, with significant positive results, were more detailed than our study, as their measurements were quantitative (enhancement ratio) or more parameters were examined. We suppose that the limited information obtained from a qualitative observation of the HBP may be responsible for the lower significant level of the results of our study, as compared to the aforementioned previous studies. Therefore, if only basic qualitative analysis of the findings in the HBP of Gd-EOB-DTPA MRI is planned, we recommend avoiding the use of Gd-EOB-DTPA MRI for diagnosis of the histological differentiation grades of HCC.
In this study, the diagnostic ability of imaging findings was the lowest for well-diff. HCC (best accuracy and AUC are 62% and 0.657, respectively). First of all, it is easy to recognize that well-diff. HCC represents an intermediate stage of HCC progression (in terms of the gradual vascular changes of destruction and formation). Therefore, the characteristics of well-diff. HCC may overlap with those of early and moderately diff. HCC, so that no unique or obvious imaging indicators could be identified in cases of well-diff. HCC. In addition, the number of well-diff. HCC lesions in our study series was relatively small (31 lesions-less than half the number of moderately diff. lesions included in this study), which could reduce the statistical power to reliably identify its characteristics.
Previous studies have shown that some vascular appearances of HCC lesions on SonoVue CEUS (such as homogeneous hyperenhancement and washout time) may be affected by the tumor size [36]. We also found that the average sizes of moderately and poorly diff. HCCs were greater than those of less advanced HCCs. This phenomenon can be explained by the possibility that the well-developed fibrous septae in well-diff. HCCs pose an obstacle to the growth of the tumors [10]. However, the non-parametric Kruskal-Wallis test to compare tumor sizes between any two groups among the four histological groups showed significant differences in the tumor size only between early HCC and other more advanced HCCs. In other words, size does not seem to play an important role in the diagnosis of early HCC. Therefore, we performed a subgroup analysis using the most valuable indicators for the diagnosis of early HCC (combination of the halo sign and PVP) according to the tumor size. We found that the diagnostic efficacy of the findings was statistically perfect in both the subgroups classified by size, with an accuracy of more than 90% and an AUC higher than 89%. These findings may suggest that the tumor size had a negligible effect on the diagnostic efficacy of the imaging indicators and patterns identified by us.
Our study had a few limitations. The first was the inadequacy of the biopsy specimens. It is generally acknowledged that histological diagnoses are far more difficult to make from biopsy specimens than from resected or autopsy specimens. Some studies have suggested that the potential problem of heterogeneous staining in biopsy specimens may reduce the diagnostic sensitivity of a biopsy as compared to resected specimens [44]. In view of this, when we recruited patients, a total of 30 patients who met all other criteria (such as performance of US and CEUS) were excluded because the biopsy did not hit the lesion, or too-little carcinoma tissue was obtained to allow a definitive histopathological diagnosis to be made. Secondly, our diagnosis of the histological differentiation grade of HCC was based on the prerequisite of a definitive diagnosis of HCC, which excluded some benign lesions such as regenerative nodules (RN). Most RNs show similar enhancement patterns (hypo, isovascularity in AP, PP, and isoechoic in PVP, respectively) to those of early HCC in Sonazoid CEUS [45]. However, RNs can be distinguished from early HCC by the central (92.3% of RN) and peripheral (97.6% of early HCC) vessel patterns in the AP [22]. In this case, if we want to make differential diagnosis of focal liver lesions (not just the histological grade of HCC), we need to pay more attention to AP.

Conclusions
To sum up, a detailed comparative analysis was made of the findings in commonly used non-invasive imaging examinations to determine the ability of individual findings or a combination of findings to allow diagnosis of the different histological differentiation grades of HCC. Most of the indicators produced positive results-especially a combination of an absence of halo signs in grayscale US and isoechoic in PVP in Sonazoid CEUS showed perfect results for diagnosing early HCC. We concluded from our study that early, poorly, and moderately diff. HCC according to our four classification method could be accurately or properly diagnosed independently or by a combination of certain patterns of image indicators (mainly Sonazoid CEUS). However, a proper diagnostic method for well-diff. HCC should be further explored. The promising results of our research will certainly be an indispensable step in towards a better application to clinical practice of our histological classification method.