Model Construction for Predicting Preoperative Microvascular Invasion of Hepatocellular Carcinoma by Gadoxetic Acid Disodium-Enhanced Magnetic Resonance Imaging Combined with Serology

Introduction

Primary liver cancer mainly includes hepatocellular carcinoma (HCC), intrahepatic cholangiocarcinoma, and mixed hepatocellular-cholangiocarcinoma, which are characterized by 3 distinct pathological types. Among these, HCC is the most prevalent type and is characterized by strong invasiveness and a high recurrence rate [1–3]. Hepatitis B virus, hepatitis C virus, excessive alcohol consumption, and metabolic disorders are among the primary risk factors for HCC. In the early stages of liver cancer, symptoms and signs are often inconspicuous. As the tumor progresses to the intermediate and advanced stages, symptoms such as hepatic pain can emerge, along with clinical signs, including hepatomegaly, jaundice, and ascites. Liver cancer with necrosis or hemorrhage can present as low-density or high-density lesions on imaging under plain scans. Due to its abundant arterial blood supply, compared with normal liver tissue, it typically exhibits the characteristic enhancement pattern of “fast in and fast out” during contrast-enhanced computed tomography (CT). According to the latest statistics, primary liver cancer ranks sixth in incidence and third in mortality globally, of which HCC accounts for over 75% of cases [4,5]. The pathogenesis of HCC is complex and affected by multiple factors, such as viral infection, genetic predisposition, and lifestyle. The worldwide incidence of HCC is increasing annually, posing a serious threat to human health.

Currently, radical resection is widely acknowledged as the primary treatment option for HCC [6]. Nevertheless, the prognosis of patients with HCC remains poor even after radical surgery, largely due to the high recurrence rate, which poses a bottleneck in HCC treatment. The recurrence rate of HCC can be as high as 70% within 5 years after surgical resection and approximately 35% after liver transplantation [7–9]. Microvascular invasion (MVI) is one of the important risk factors for early recurrence of HCC after hepatectomy or liver transplantation [10–12]. Since MVI can be diagnosed only by postoperative pathology, preoperative diagnosis is relatively difficult. Due to the heterogeneity of tumors, preoperative pathological biopsies have been shown to be highly inaccurate and can potentially pose a risk of tumor dissemination. Currently, the preoperative prediction of MVI relies mainly on the application of gadoxetic acid disodium (Gd-EOB-DTPA)-enhanced magnetic resonance imaging (MRI). In recent years, the relationship between tumor and inflammatory indicators has become a research hotspot, and some scholars have found a link between HCC MVI and inflammatory factors [13,14]. In the present study, the preoperative Gd-EOB-DTPA-enhanced MRI features, preoperative clinical data, and pathological data of 120 patients with HCC admitted to the Affiliated Huai’an Hospital of Xuzhou Medical University from January 2017 to December 2023 were retrospectively analyzed, with the aim of investigating the predictive value of preoperative Gd-EOB-DTPA-enhanced MRI combined with serological inflammatory indicators on MVI in patients with HCC.

Material and Methods

GENERAL INFORMATION:

A total of 120 patients with HCC were enrolled in this study and divided into a training group and a validation group in a ratio of 7: 3, of which 95 were men and 25 were women. The age of the patients ranged from 35 to 83 years, with a mean age of 58.78±10.58 years. All patients underwent preoperative examination in our hospital and were operated on by the same team of surgeons. Postoperative pathology confirmed HCC. Hepatitis B virus surface antigen (HBsAg) positivity was high in the study population, with 82 cases being positive and 38 cases being negative (Figure 1).

INCLUSION AND EXCLUSION CRITERIA:

The inclusion criteria were as follows: (1) postoperative histopathological diagnosis of HCC and R0 resection; (2) no distant metastasis; (3) patients first diagnosed at our hospital and receiving no other anti-tumor treatments such as surgical therapy, trans-arterial chemoembolization, radiation therapy, and immunotherapy prior to enrollment; and (4) relevant laboratory and imaging tests performed within 4 weeks after surgery.

The exclusion criteria were as follows: (1) incomplete clinical data or laboratory examination indicators; (2) poor image quality affecting analysis; (3) concurrent systemic malignant tumors; (4) visible vascular invasion; and (5) Child-Pugh grade C.

EXAMINATION METHODS:

A GE Discovery MR750w 3.0T superconducting MR scanner (GE Healthcare, Milwaukee, Wisconsin, USA) with a 16-channel abdominal coil was used to scan the whole liver and spleen. Enhanced MR examination was performed using the LAVA sequence, with the following scanning parameters: TR 4.2 ms, TE 1.7 ms, layer thickness 5.0 mm, layer spacing 2.5 mm, field of view 40×40 cm, and number of excitations 0.68. The examination was performed after fasting for 6 to 8 h, and the gadoxetic acid disodium, a contrast agent, was injected intravenously through the elbow joint at a dose of 0.025 mmol/kg and an injection flow rate of 2.0 mL/s. After injection, the catheter was flushed with an additional 20 mL of physiological saline. Scans were performed in the pulsatile, portal, delayed, and hepatocellular phases, with intervals of 20 s, 40 s, 3 min, 10 min, and 20 min, respectively.

IMAGING ANALYSIS:

Preoperatively, 2 senior radiologists from the imaging department of our hospital classified liver tumors into infiltrative and non-infiltrative based on MRI imaging, MVI-related imaging features, and morphology, mainly in the Gd-EOB-DTPA hepatobiliary phase (HBP), by combining with other sequences, observing their axial images, and combining with the coronal position. Infiltrative tumors are defined as masses that are irregular in shape, lack well-demarcated boundaries without obvious envelope, and have “cut-through” changes with the surrounding normal liver tissues. Periarterial high-signal ring refers to the regular or irregular high-signal ring around the tumor edge in the arterial phase, with relatively low signal, compared with the liver tumor. Hepatobiliary tumor signal refers to the signal of the hepatobiliary liver tumor relative to the surrounding normal liver parenchyma. If the signal intensity in the tumor is lower than that in the surrounding liver parenchyma, it is defined as low signal; if the signal in the tumor is high or heterogeneous, it is defined as high or mixed signal. Peritumoral hypointensity in the HBP is defined as an irregular low-signal area in the peripheral region of the liver tumor in HBP (Figure 2). In the event of any discrepancies between the 2 senior radiologists, the films were reviewed by a third senior radiologist, and the final decision was reached by all 3 radiologists. An interobserver agreement test was performed at the end of the review.

STATISTICAL ANALYSIS:

SPSS 22.0 statistical software was used for data analysis. Normally distributed variables were presented as mean±standard deviation, and the comparisons between groups were performed using the least significant difference t test. Skewed variables were expressed as median (interquartile range), and the comparisons between groups were performed using the Mann-Whitney U test. Categorical variables were presented as number of cases (percentage), and the comparisons between groups were made using the chi-square and Fisher exact probability tests. Univariate and multivariate logistic regression analyses were conducted for the MVI (−) and MVI (+) groups in the training cohort, and an imaging-based prediction model was constructed using the resulting independent risk factors. ROC curve analysis was performed, and the nomograms, calibration curves, and decision curves of the prediction model were generated by the R rms package for the study. The McNemar test and Kappa test were used for consistency assessment. A Kappa value ≥0.75 indicated very good consistency, a Kappa value <0.40 indicated poor consistency, and 0.40≤ a Kappa value <0.75 indicated moderate consistency. P<0.05 was considered statistically significant.

Results

CLINICAL AND IMAGING FEATURES OF PATIENTS WITH HCC:

Among the 120 patients, 95 were men, accounting for a relatively high proportion (79.2%). The patients’ age was basically distributed around 58 years old, with a small age difference between patients. The percentage of patients that were HBsAg-positive was higher than the general percentage, at 68.3%. The serological indicators varied greatly between patients, and the percentage of non-infiltrative tumors in imaging features was relatively high, at 72.5% (87 cases). The overall percentages of intratumoral arterial imaging features in the arterial phase of the tumor and high or mixed signals in the hepatobiliary phase of the tumor were 41.7% (50 cases) and 45.0% (54 cases), respectively, which were higher than other imaging features. No significant differences were observed in the baseline clinical characteristics between the training cohort and the validation cohort (Table 1). K test analysis of the 2 imaging radiologists’ observations showed that the results were greater than 0.75, indicating a good consistency with the obtained data.

UNIVARIATE ANALYSIS FOR PREDICTING MVI:

The results of the univariate analysis (Table 2) indicated that alpha-fetoprotein (AFP), C-reactive protein to albumin ratio (CAR), tumor signal in HBP, and peritumoral hypointensity in HBP were correlated with MVI, with P values of 0.026 (OR 1.002, 95% CI 1.000–1.005), 0.004 (OR 26.971, 95% CI 5.568–30.568), 0.049 (OR 2.459, 95% CI 1.005–6.016), 0.011 (OR 15.667, 95% CI 1.881–130.456), respectively. The rest of the indicators were not statistically significant with MVI, among which the tumor diameter might be non-statistically significant due to the small sample size.

MULTIVARIATE ANALYSIS FOR PREDICTING MVI:

The results of multivariate logistic regression model analysis are shown in Table 2: AFP (OR 1.002, 95% CI 1.000–1.005, P=0.052); CAR (OR 35.312, 95% CI 3.064–43.293, P=0.029); tumor signal in HBP (OR 3.559, 95% CI 1.206–10.499, P=0.021); and peritumoral hypointensity in HBP (OR 15.296, 95% CI 1.741–134.358, P=0.014).

ROC CURVE ANALYSIS OF PREDICTIVE EFFICACY:

In this study, the predictive efficacy of CAR, tumor signal in HBP, and peritumoral hypointensity in HBP on HCC MVI when applied alone or in combination was analyzed by the ROC curve (Figure 3). The results showed that the AUC of the 3 factors when predicted independently was 0.674 (95% CI 0.558–0.790), 0.608 (95% CI 0.484–0.731), and 0.615 (95% CI 0.489–0.740), and when the tumor imaging features were applied in 2-by-2 and 3-by-3 combinations with CAR, the AUC was 0.718 (95% CI 0.608–0.829), 0.748 (95% CI 0.641–0.854), and 0.789 (95% CI 0.686–0.892), respectively. CAR, tumor signal, and peritumoral hypointensity in HBP had sensitivities of 54.2%, 68.8%, and 97.9%, and specificities of 80.6%, 52.8%, and 25.0%, respectively; the sensitivities of tumor imaging features and CAR for 2-by-2 and 3-by-3 combinations were 60.4%, 54.2%, and 87.5%, respectively, and the specificities were 72.2%, 63.9%, and 86.1%, respectively. Compared with the other models, the model constructed by combining the 3 indicators had better predictive efficacy, with the largest AUC of 0.789, which was generally better in assessing MVI in terms of sensitivity and specificity. In the validation group, a comparison of the area under the ROC curve between the combined model and the single factor revealed that the third combined model exhibited the best performance, with an area under the curve of 0.897, which was consistent with the findings in the training group.

COLUMN LINE GRAPH, CALIBRATION CURVE, AND DECISION CURVE:

The results of multivariate logistic analysis indicated that CAR, tumor signal in HBP, and peritumoral hypointensity in HBP should be included in the column line graph for predicting HCC MVI (Figure 4). The apparent and bias-corrected curves for the training group were in the same direction and close to the reference line when the calibration curve was plotted, indicating good model consistency (Figure 5). The data for the validation group were included in the same model, with predicted and actual curves close to each other, which indicated that the column-line graph model constructed in this study had good consistency (Figure 6). Decision curve analysis showed curves for the training and validation groups were 0.2 to 0.9 for threshold probability, indicating that clinical interventions based on the diagnostic results of column-line graph would provide greater clinical benefits for patients (Figure 7).

Discussion

The improvement of preoperative predictive efficacy for HCC MVI aids in the selection of treatment options for patients, especially for those with surgical indications, in which the choice of surgical approach is critical. Many scholars believe that when patients have MVI, anatomic liver resection has a greater chance of achieving radical resection, increasing the tumor-free survival rate and overall survival rate, and improving patient prognosis [15,16]. At present, the diagnosis of HCC MVI mainly relies on the criterion standard of postoperative pathology. With the development of the last decade or so, there is an increasing number of studies on the prediction of HCC MVI through imaging; especially, the emergence of Gd-EOB-DTPA contrast agent has pushed MRI to the climax of imaging for HCC MVI. If prediction is performed solely by Gd-EOB-DTPA-enhanced MRI, its predictive specificity is high, but its sensitivity profile is not favorable. Under the current research hotspot of inflammatory factors and tumors, it has been shown that there is a link between CAR and the treatment and prognosis of various malignant tumors [17–20]. The inflammatory factors in serology are combined with imaging to predict HCC MVI, in order to establish a new prediction model and improve the predictive efficacy. The results of univariate logistic analysis in this study showed that CAR, AFP, tumor signal in HBP, and peritumoral hypointensity in HBP were risk factors for MVI. The results of multivariate logistic analysis showed that CAR, tumor signal in HBP, and peritumoral hypointensity in HBP were independent risk factors for MVI.

Inflammatory factors play a critical role in the development of many malignant tumors, and the tumor mesenchyme represents a chronic inflammatory microenvironment where inflammatory factors promote tumor cell proliferation, invasion, and metastasis [21]. C-reactive protein is a common inflammatory indicator in serology, which can reflect systemic and local inflammation, as well as evaluate the severity and prognosis of liver tumors [22,23]. The CAR, or C-reactive protein to albumin ratio, reflects not only the inflammatory status of patients with tumors, but also their nutritional and immune levels. Logistic regression analysis in this study showed that the CAR was an independent risk factor for MVI, and the column line graph showed that the higher the ratio, the higher the probability of MVI risk in patients. Ma et al [24] found that HCC MVI patients with high CAR levels had poorer prognosis than did those with low CAR levels. Some scholars have reported that CAR, as a novel systemic inflammatory indicator, can predict preoperative MVI in HCC, and is more efficient in predicting MVI in single small HCC when CAR ≥0.03 [25].

The present study showed that tumor signal in HBP was an independent risk factor for predicting MVI in HCC, and a high or mixed signal in HBP of the tumor suggested an increased probability of MVI in HCC, predicting poorer prognosis. Gd-EOB-DTPA, a specific MR contrast agent, enters the liver due to its fat-soluble ethoxylated phenylmethyl (EOB) cells, presenting specific HBP imaging features with low signal due to the inability of tumor cells to uptake Gd-EOB-DTPA, and high signal due to the ability of normal hepatocytes to uptake Gd-EOB-DTPA [26], making it easier to clarify the nature of hepatic occupancy, compared with other contrast agents. In this study, the proportion of mixed signals in the tumor signals in HBP of HCC MVI patients was higher than that in patients with completely high signals. In addition, some studies have shown that inhomogeneous mixed signals in HBP of tumors have a higher degree of malignancy and poorer prognosis than those with completely low or high signals [27].

Furthermore, in the HBP of Gd-EOB-DTPA MRI in this study, in addition to tumor signal, peritumoral hypointensity was an independent risk factor for MVI, which was lower than that of other normal liver tissues. It has been indicated that the main mechanism is that MVI leads to damage of tumor margin hepatocytes due to hemodynamic alterations, and their inability to uptake Gd-EOB-DTPA usually manifests as low signal [28–30]. In the present study, peritumoral hypointensity in HBP had a greater effect on MVI, as shown by the line graphs, and many scholars have shown that there is a high correlation between peritumoral hypointensity in HBP and MVI in HCC [28,31].

Conclusions

In summary, Gd-EOB-DTPA-enhanced specific HBP MRI and novel inflammatory indicators are beneficial for the prediction of preoperative MVI, as well as for assisting preoperative clinical decision-making and improving patient prognosis. CAR, tumor signal in HBP, and peritumoral hypointensity in HBP are independent risk factors for MVI in HCC, and the predictive model constructed by combining the 3 indicators has better predictive efficacy than does a single indicator. The present study is a retrospective study with a small sample size and certain limitations. However, with in-depth study of inflammation and tumor indicators, as well as in-depth study of Gd-EOB-DTPA MRI and other imaging, the combination of 3 indicators will make the preoperative prediction of MVI increasingly accurate.