Value of gadoxetic acid-enhanced MRI for microvascular invasion of small hepatocellular carcinoma: a retrospective study

Background The objective of this study was to analyze the accuracy of gadolinium–ethoxybenzyl–diethylenetriamine penta–acetic acid enhanced magnetic resonance imaging (Gd–EOB–DTPA–MRI) for predicting microvascular invasion (MVI) in patients with small hepatocellular carcinoma (sHCC) preoperatively. Methods A total of 60 sHCC patients performed with preoperative Gd–EOB–DTPA–MRI in the Harbin Medical University Cancer Hospital from October 2018 to October 2019 were involved in the study. Univariate and multivariate analyses were performed by chi–square test and logistic regression analysis. The sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of Gd–EOB–DTPA–MRI were performed by receiver operating characteristic (ROC) curves. Results Univariate analysis indicated that alanine aminotransferase (≥ 39.00U/L), poorly differentiated pathology, and imaging features including grim enhancement, capsule enhancement, arterial halo sign and hepatobiliary features (tumor highly uptake, halo sign, spicule sign and brush sign) were associated with the occurrence of MVI (p < 0.05). Multivariate analysis revealed that rim enhancement and hepatobiliary spicule sign were independent predictors of MVI (p < 0.05). The area under the ROC curve was 0.917 (95% confidence interval 0.838–0.996), and the sensitivity was 94.74%. Conclusions The morphologies of hepatobiliary phase imaging, especially the spicule sign, showed high accuracy in diagnosing MVI of sHCC. Rim enhancement played a significant role in diagnosing MVI of sHCC.

invasion (MVI) after the radical operation, causing an unsatisfactory prognosis [4]. Therefore, predicting MVI of sHCC before the surgical operation could guide the clinician to choose proper strategies, then improving outcomes of patients.
Gadolinium-ethoxybenzyl-diethylenetriamine penta-acetic acid (Gd-EOB-DTPA) is well known as a perfect liver-specific contrast agent. Due to hepatic cell uptake, it can distinguish abnormal lesions from normal liver parenchyma easily and improve small lesions′ detection rate [5,6]. Hence, gadoxetic acid is considered as the most critical contrast medium for diagnosing and detecting HCC. Besides, the studies that predicting MVI ofsHCC by Gd-EOB-DTPA enhanced magnetic resonance imaging (Gd-EOB-DTPA-MRI) before surgery were relatively immature. Some studies had attempted to predict MVI preoperatively by Gd-EOB-DTPA-MRI [7][8][9][10][11][12][13][14]. Peritumoral hyper enhancement on the arterial phase [15] and hypouptake on hepatobiliary phase (HBP)are known as prominent risk factors of MVI [16,17]. However, to the best of our knowledge, few systematic study assessed the efficacy of Gd-EOB-DTPA-MRI in predicting MVI of sHCC. Our study intergrated the imaging and clinicopathological features and aimed to find effective marker sin predicting MVI of sHCC preoperatively.

Patients
A total of 130 HCC patients who received curative hepatic resection at our hospital between October 2018 and October 2019 were enrolled in this study. The inclusion criteria for our study were as follow: (a) HCCs′ diameters were equal to or less than 3 cm in maximum; (b) Patients underwent preoperative Gd-EOB-DTPA-MRI within one month before surgery; (c) There were no grossly vascular tumor thrombosis or extrahepatic metastasis on preoperative imaging evaluation;(d) There were no preoperative treatments; (e) Full histologic description was available in the pathologic reports and imaging quality adequate for analysis; (f ) All malignant nodules were involved for analysis in patients with multiple sHCCs. Finally, 62 nodules in 60 patients were included in the present retrospective study (Fig. 1).This study was approved by the Ethics Review Board of Harbin Medical University. The requirement of informed consent from the patients was waived because of the retrospective design of this study, and patients′ information was protected.

Image analysis
Preoperative MRI images were retrospectively evaluated using a Picture Archiving and Communication System (PACS; GE Medical Systems Integrated Imaging Solutions, Mt. Prospect, IL, USA) with an optimal window setting adjustment in each case. Two experienced abdominal radiologists interpreted the imaging analysis (with 12 and 20 years of experience in HCC, respectively). They were blind to clinical, pathological and MVI information.
Our study evaluated the imaging features of each sHCC and focused on the following imaging features.

Non-enhancement imaging
(a) Mosaic architecture: it is anancillary feature of Liver Imaging Reporting And Data Systemthat is favoring HCC in particular (American College of Radiology (ACR). Liver Reporting & Data System (LI-RADS). ACR website.www.acr.org/Clini cal-Resou rces/Repor ting-and-Data-Syste ms/LI-RADS.). Mosaic architecture refers to the difference present within mass of randomly distributed internal nodules or compartments differing in enhancement, attenuation, intensity, shape, and size and often separated by fibrous separations [18]. It is characteristic of a heterogeneous signal on T2WI [19]; (b) Intralesional fat: the chemical shift of intralesional fat cause an area shows significantly lower signal intensity (SI) on opposed-phase imagesT1WI compared with the SI in-phase images [20], which confirmed the presence of steatosis [21]; (c) intratumor hemorrhage: with ahyper-SI on unenhanced T1WI and a hypo-SI on T2WI [21]; (d) Iso/Hyperintense SI (T1WI):isointense (hyper-isointense) SI in the lesion on T1WI; (e) T2-weighted and diffusion-weighted imaging (T2-DW) mismatch: T2-DW mismatch mainly means a morphological mismatch, with a larger mass on thediffusion weighted imaging (DWI) images than T2WI images. Because of the image distortion on the DWI images, it was not accurate that only compare the lesion diameter or area on DWI with T2WI. So, we added the other requirement. The mismatch region's intensity on DWI was lower than the tumor itself but higher than the liver parenchyma. These conditions are the same as the previous study [22] (Figs. 2, 3); (f ) Morphology: it was assessed on T2WI and categorized as round, lobulated and irregular shape.

Enhancement without HBP imaging
(g) AP hyper-enhancement: it was referred to AP enhancement that unequivocally greater than the background of hepatic parenchyma rather than a rim structure; (h) Washout: it was based on AP hyperenhancement and followed by a lower SI than the normal hepatic parenchyma on PVP and/or EP [23]; (i) Rim enhancement: it was defined as irregular ring-like enhancement with relatively hypovascular central areas in the AP; (j) Capsule enhancement: it was assessed during EP and defined as thin, linear and enhanced structure surrounding the tumor; (k) AP halo sign: it was described as an irregular and ring-like enhancement sign that adjacent to tumor border.

HBP imaging
(l) HBP high uptake: it shows an increased tumor uptake of contrast agents and as lightly high SI in the lesion area [24,25]. (m) HBP halo sign: it was a hypo-intensity ring around tumors [15]. (n) HBP morphology: it was categorized on T2WI images. (o) HBP spicule sign: it was a thick, blurred, and pseudopodal structure. (p) HBP brush sign: it was the edges of the tumor and appeared as a blurry, small corner projection (Fig. 4).

Pathology
The diagnosis of MVI was based on histologic specimens obtained from surgical resection in all patients. Based on the World Health Organization classification system, an experienced pathologist (with nine years of experience) who was blind to all clinical and MRI results confirmed the histologic diagnosis and assessed histologic grade, vascular invasion, and tumor diameter.
According to the classification proposed by the General Rules of the Clinical and Pathological Study of Primary Liver Cancer in Japan [26], we classified the tumor grade into highly, moderately and lowly differentiated. Besides, the predominant grade was assigned when grades coexisted. The tumor diameter was defined as the maximum diameter of the resected tumor specimen.
We consider that pathological materials should combine with the MRI findings. Once patients with MVI areassessed before surgery, the scope of surgery or pathological materials should be expanded in the area where the MVI is suspected.

Statistical analysis
We used SPSS software (version 22.0, IBM) to analyze all data. Categorical variables were compared by the chi-square test and followed by a multivariate logistic regression analysis when the variables were significant in the chi-square test. Then we divided all features into Non-HBP (non-enhancement and enhancement without HBP), HBP and the combination of those two phases. We performed the logistic regression analysis to assess potential imaging predictors for MVI in different phases. Variables with p < 0.05 in the uni variate analysis were applied to multivariate logistic regression analysis. Features with p value less than 0.05 were considered statistically significant. A receiver operating characteristic (ROC) curves were performed to evaluate the diagnostic ability of each phase and their combination. The value of sensitivity (SE), specificity (SP), accuracy, positive predictive value (PPV), and negative predictive value (NPV) in these three parts were also calculated. Additionally, to test the generalizability of the tested and the inter observer variability of the radiological features, we assessed the intra class correlation coefficient test (ICC) (ĸ = 0.00-0.20, poor agreement; ĸ = 0.21-0.40, fair agreement; ĸ = 0.41-0.60, moderate agreement; ĸ = 0.61-0.80, good agreement; ĸ = 0.81-1.00, excellent agreement) [27].

Clinicopathological features results
Among 62 sHCC nodules (60 patients), 19 nodules (19 patients) had MVI, while 43 nodules (41 patients) had no MVI. The study included 47 men and 13 women, with a median age of 55 . Comparisons of clinical characteristics between sHCC with and without MVI were summarized in Table 1. Alanine aminotransferase (ALT) was higher in sHCC patients with MVI than those without MVI. Concerning histologic features, sHCC with MVI showed worse tumor differentiation than those without MVI (p < 0.05).

Univariate analysis of clinical, pathological and imaging features
Among MRI features showed in Table 2, on nonenhancement phase, T2-DW mismatch was significantly more frequent in sHCC with MVI than those without MVI (p < 0.05). On enhancement phase without HBP, rim enhancement, capsule enhancement and AP halo sign were associated with MVI (p < 0.05). All imaging features including high uptake, halo sign, spicule sign and brush sign on HBP were more common in sHCC with MVI than those without MVI (p < 0.05). Tumor morphology of lobulated and irregular shape in sHCC with MVI was significantly more frequent than that of sHCC without MVI (p < 0.05).Besides, the round shape showed more common in sHCC without MVI than those with MVI (p < 0.05). Themorphology features of sHCC are presented in Fig. 4. A round shape of sHCC with MVI accounts for6.80% (3/44),a lobulated shape accounts for 45.80%(11/24), an irregular shape accounts for 80.00%(4/5), a spicule sign accounts for 80.00%(12/15) and a brush sign accounts for 66.70%(8/12), respectively.

Univariate and multivariate logistical regression analyses of clinical, pathological and imaging features
In univariate logistical regression analysis, ALT, rim enhancement and spicule sign were associated with MVI (p < 0.05). And the association remained significant for rim enhancement and spicule sign in multivariate logistical regression analysis (p < 0.05), which were summarized in Table3.

Evaluate the diagnostic ability of gadoxetic acid-enhanced MRI
We analyzed the diagnostic efficacy of two significant imaging phases and their combination for predicting MVI (Fig. 5, Table 5).When two of these imaging phases were combined, the area under the ROC curve (AUROC) was higher than the non-HBP phase or HBP separately.
The analysis of SE, SP, accuracy, PPV, and NPV were presented in Table 5.Compared to non-HBP phase or HBP, the combination of these two imaging phases showed significantly higher SE, accuracy and NPV. Consequently, the combination of these two imaging phases can predict MVI more accurately compared with them separately.

Discussion
The purpose of this research was to predict MVI in sHCC by imaging features, clinical features and pathological features, providing a direction for clinicians to choose a superior treatment strategy then improving patients′ survival.
Our research used a special contrast agent and achieved satisfactory results. Gd-EOB-DTPA, a hepatocyte-specific contrast agent, has dual enhanced information which not only exhibits a multi-phase dynamic enhancement similar to extra cell contrast medium by reducing the T1 relaxation time of tissues but also can acquire high uptake images of normal liver parenchyma after administrating the agent for 20 min. Therefore, this particular agent can deliver more information about the HCC from AP, PVP, EP and HBP. It is more effective than traditional MRI in diagnosing early HCC and sHCC [28], especially in the detection of tiny lesions [29]. Meanwhile, because of the background of high uptake of normal liver tissue on HBP, the morphology of the tumor appears much clearly [30].
Gd-EOB-DTPA contrast agent has many advantages in hepatic MRI examination, however, transient respiratory motion artifacts result in a diagnostic limitations about the observation of AP information. Finding more significant features to improve MVI diagnosis by observing the morphology of the lesions during the HBP is extremely necessary. Up to date, studies discuss the features that associated with the occurrence of MVI in sHCC were rare. Xu etal [31] proved that the high ADC value and irregular circumferential enhancement were independent predictors of MVI, which was consistent with our research results. Kim and his colleagues showed that all sHCCs with MVI were characterized by a typical dynamic pattern, hyper intensity on T2WI, DWI, atypical dynamic pattern. Instead, the size of diameter was less than 1 cm indicated the absence of MVI [32].    [14]. However, our research did not get the same results, and we regarded a relatively higher ALT value as a significant feature in MVI prediction (p < 0.05). In our study, by comparing clinic pathological features and imaging features of non-HBP and HBP, we found that the HBP features provided us more insights for MVI diagnosis ( Table 2). All of the characteristics we discovered were significantly different between MVI-positive and MVI-negative group. About 3/5 of characteristics (HBP tumor morphology, spicule sign and brush sign) were morphology features. Up to date, several studies proved that the irregular tumor shape was associated with the presence of MVI [33,34]. In particular, our research is the first attempt to classify the irregular shape of sHCC in detail. As a result, we should pay attention to the spicule sign that is contained in most MVI-positive patients (12/19, 63.20%), and few (3/43, 7.00%) in MVI-negative sHCCs. Sign of spicule between the lobe and burr represents tumor infiltration into the base of the lobes. As an independent risk factor, our research showed a better performance in diagnosing MVI than irregular shape and brush features. The spicule sign was quoted from lung cancer and it′s detection rate was approximately 90.00%, which was a central differential diagnostic marker of pulmonary nodules [35] and predicted poor biological behavior [36]. According to our cases, it is also appropriate for the judgment of the biological behavior of sHCC. Besides, it should be noted that the morphological change is more easily discovered and acceptable for the clinician than other complex enhancement and signal features.
Except for spicule sign on HBP images, rim enhancement was another independent risk factor for diagnosing MVI on non-HBP images. There were 9 cases with MVI in all cases of rim enhancement (9/13, 69.23%), and 12 cases with MVI in all cases of none capsule enhancement (12/16, 75.00%). Rim enhancement on AP reflected the internal tissue of fiber necrosis, liquefaction or calcification caused by the insufficient blood supply. We speculated obstruction of surrounding from MVI was likely to further exacerbate the blood loss, intensifying internal necrosis as well as margin enhancement.
Finally, we divided all features into two phases, non-HBP (non-enhancement and enhancement without HBP)and HBP alone. And we compared the diagnostic capability of each phase and the combination phase. The combining phase group achieved the highest AUROC (0.917) than the two phases alone. Meanwhile, HBP image features give the best result inaccuracy and SP, which was entirely credible in predicting MVI in sHCC. To the best of our knowledge, the main reasons that the SE of the combined phase images and SP of HBP images (> 90%) in our research might be explained as follows. Firstly, Gd-EOB-DTPA-MRI can accurately discriminate cancer boundaries. Moreover, the distinction in  signal between lesion tissues and surrounding normal liver parenchyma is more apparent in the HBP of Gd-EOB-DTPA-MRI than conventional contrast agents [37,38], which contributes to the highest SP in HBP than non-HBP. Secondly, owing to the high performance of radiomics in the stable calculation, repeatability, indefatigability, and no interference of human subjectivity. Our research made a much higher SE and SP than those ever reported [39][40][41][42]. Thirdly, we extracted almost the whole sHCC features on HBP, including the three-dimensional features (e.g., morphology and smoothness), making the study much more credible and representative. Moreover, we found that the NPV of non-HBP images was very low, which suggested a high possibility of missed diagnosis. Thus, we suggested that the characteristics of HBP images alone as predominant indicators, especially morphologic features, which could increase radiologist confidence in diagnosing sHCC with MVI. However, our study had the following limitations. Firstly, this was a retrospective study that may have included selection bias. Secondly, the number of the sample was insufficient, which may produce errors in data analysis. Thirdly, it was a single center study, which may also cause sample selection bias. Therefore, it is worthwhile to conduct further research to verify our results.

Conclusions
This study suggested that Gd-EOB-DTPA-MRI was recommended as a routine preoperative examination for sHCC to improve the accuracy of MVI diagnosis. The morphologies of HBP imaging, especially sign of spicules, showed high accuracy in diagnosing MVI of sHCC.