Comparison contrast-enhanced CT with contrast-enhanced US in diagnosing combined hepatocellular-cholangiocarcinoma: a propensity score-matched study

Objectives To develop and compare noninvasive models for differentiating between combined hepatocellular-cholangiocarcinoma (cHCC-CCA) and HCC based on serum tumor markers, contrast-enhanced ultrasound (CEUS), and computed tomography (CECT). Methods From January 2010 to December 2021, patients with pathologically confirmed cHCC-CCA or HCC who underwent both preoperative CEUS and CECT were retrospectively enrolled. Propensity scores were calculated to match cHCC-CCA and HCC patients with a near-neighbor ratio of 1:2. Two predicted models, a CEUS-predominant (CEUS features plus tumor markers) and a CECT-predominant model (CECT features plus tumor markers), were constructed using logistic regression analyses. Model performance was evaluated by the area under the curve (AUC), sensitivity, specificity, and accuracy. Results A total of 135 patients (mean age, 51.3 years ± 10.9; 122 men) with 135 tumors (45 cHCC-CCA and 90 HCC) were included. By logistic regression analysis, unclear boundary in the intratumoral nonenhanced area, partial washout on CEUS, CA 19-9 > 100 U/mL, lack of cirrhosis, incomplete tumor capsule, and nonrim arterial phase hyperenhancement (APHE) volume < 50% on CECT were independent factors for a diagnosis of cHCC-CCA. The CECT-predominant model showed almost perfect sensitivity for cHCC-CCA, unlike the CEUS-predominant model (93.3% vs. 55.6%, p < 0.001). The CEUS-predominant model showed higher diagnostic specificity than the CECT-predominant model (80.0% vs. 63.3%; p = 0.020), especially in the ≤ 5 cm subgroup (92.0% vs. 70.0%; p = 0.013). Conclusions The CECT-predominant model provides higher diagnostic sensitivity than the CEUS-predominant model for CHCC-CCA. Combining CECT features with serum CA 19-9 > 100 U/mL shows excellent sensitivity. Critical relevance statement Combining lack of cirrhosis, incomplete tumor capsule, and nonrim arterial phase hyperenhancement (APHE) volume < 50% on CECT with serum CA 19-9 > 100 U/mL shows excellent sensitivity in differentiating cHCC-CCA from HCC. Key points 1. Accurate differentiation between cHCC-CCA and HCC is essential for treatment decisions. 2. The CECT-predominant model provides higher accuracy than the CEUS-predominant model for CHCC-CCA. 3. Combining CECT features and CA 19-9 levels shows a sensitivity of 93.3% in diagnosing cHCC-CCA. Graphical Abstract Supplementary Information The online version contains supplementary material available at 10.1186/s13244-023-01576-6.


Introduction
Combined hepatocellular-cholangiocarcinoma (cHCC-CCA) accounts for 0.4-4.2% of primary liver cancer cases and demonstrates hepatocytic and biliary differentiation in the same tumor [1].In routine practice, a marked overlap in clinical conditions has been observed between cHCC-CCA and hepatocellular carcinoma (HCC) [2,3].
However, the treatment strategies for cHCC-CCA and HCC differ.For example, liver transplantation has been accepted as an effective curative-intent treatment option for HCC, but it is not recommended for cHCC-CCA because of frequent recurrence (54% at 5 years) and suboptimal long-term survival (41% at 5 years) [4][5][6].Lymph node dissection is recommended for resectable cHCC-CCA, while it is not routinely performed for HCC [1,7].Local treatment and systematic therapy are established treatment options for unresectable HCCs [8,9], but whether they have therapeutic benefits for unresectable cHCC-CCAs is controversial [6,10].Thus, the accurate differential diagnosis between cHCC-CCA and HCC is critical for appropriate therapeutic decision-making.
Contrast-enhanced ultrasound (CEUS) and contrastenhanced computed tomography (CECT) are two of the main imaging modalities for diagnosing liver tumors [11,12], and each of these modalities has different imaging principles and advantages.CEUS is performed with pure blood contrast and can continuously evaluate macro-and microvascular perfusion within tumors [13].CECT uses a small molecule contrast agent that can assess the entire liver during a standard examination and has demonstrated clear advantages over CEUS in detecting tumors and extrahepatic lymph metastases [14].Previous studies have explored the potential of combining laboratory results with CEUS or CECT for differentiating cHCC-CCA from HCC but have yielded suboptimal diagnostic performances (sensitivity for cHCC-CCA: 32.5 to 74.4%) [2,[15][16][17][18].No studies have compared the diagnostic efficacy of CEUS and CECT in differentiating cHCC-CCA from HCC.It is also uncertain which imaging examination is best for diagnosing probable HCC/cHCC-CCA in high-risk patients.
Here, we aimed to develop diagnostic models integrating clinical and readily accessible CEUS and CECT features to differentiate between cHCC-CCA and HCC in a propensity score-matched study and to compare the two models.

Patients
From January 2010 to December 2021, patients who underwent curative-intent liver resection for surgically proven HCC or cHCC-CCA were consecutively enrolled.The inclusion criteria were as follows: (a) pathologically proven HCC or cHCC-CCA, (b) both CECT and CEUS examinations within 1 month before surgery, and (c) chronic hepatitis B/C virus infection or cirrhosis.Patients were excluded if (a) they had received any prior antitumoral treatment, (b) key laboratory data were not available, or (c) CECT and/or CEUS images were degraded or missing.The inclusion and exclusion flowchart is shown in Fig. 1.

Imaging acquisition
The imaging acquisition recommendation and the detailed parameters are presented in eMethods 1 in Supplement 1.

Imaging analysis
All image analyses were conducted on a per-lesion basis by two ultrasonographers (K-Y.Z. and J-Y.H., with 8 and 10 years of experience in CEUS images, respectively) and two radiologists (Y.Z. and Y-D.C., with 8 and 11 years of experience in liver imaging in CECT images, respectively).All disagreements between the reviewers regarding the imaging features were resolved by consensus.For patients with multiple lesions, the largest targeted lesion was selected for feature-related analyses.

Histopathology analysis
The pathological characteristics of the lesions were retrospectively recorded according to the pathological reporting system in our hospital.These included the maximum size of the main lesion (the largest lesion in the case of multiple lesions), the Edmondson-Steiner grade of the HCC, and the HCC/intrahepatic cholangiocarcinoma (ICC)-predominant components of cHCC-CCA [28].

Statistical analysis
Propensity score-matching was performed to minimize the effect of potential selection bias and confounding factors between patients with HCC and cHCC-CCA.
The predictive models, based separately on tumor markers and CEUS and CECT features, were constructed using logistic regression analyses, and their performance was compared with that of pathology.The variables with p < 0.05 by the χ 2 test or Fisher's exact test were entered into the univariate logistic analysis, and the multicollinearity between the univariate variables was assessed using Spearman's correlation analysis and by computing the variance inflation factor (VIF).If the absolute value of the correlation coefficient (ACC) was ≥ 0.6 or the VIF was > 10 between two variables, the variable with the higher odds ratio (OR) were selected for multivariate logistic analysis.Therefore, all independent variables that were associated with cHCC-CCA in univariate analyses were input into a multivariate logistic regression model using the backward stepwise method while adjusting for the same covariates as above.The diagnostic models were illustrated as nomograms based on their correlation coefficients in the multivariate logistic analysis.
Model discrimination was assessed by computing the area under the receiver operating characteristic curve (AUC) value and compared using the DeLong test.Model calibration was evaluated by the Hosmer-Lemeshow (H-L) test and calibration curves.The McNemar test was used to compare pairwise sensitivities, specificities, and accuracies of the two diagnostic models.The subgroup comparison of the diagnostic efficacy between CEUS and CECT was also evaluated for smaller lesions, with a diameter of 5 cm.
All statistical analyses were performed with the R software (R Foundation for Statistical Computing, version 3.2.5, http:// www.r-proje ct.org/) and MedCalc (version 10.4).A two-tailed adjusted p < 0.05 was statistically significant.
The key clinical features of the patients are summarized in Table 1.

Imaging features and interrater agreement
Based on the CEUS LI-RADS classification, 46.6% and 37.8% of cHCC-CCA patients were classified as LR-4/5 and LR-M, respectively; for HCC, 65.6% and 20% were classified as LR-4/5 and LR-M, respectively.Based on the CECT LI-RADS classification, 44.5% and 51.1% of cHCC-CCA patients were classified as LR-4/5 and LR-M, respectively, compared with 76.7% for LR-4/5 and 12.2% for LR-M among the HCC patients.
On CEUS, the following features were more frequent in patients with cHCC-CCA than in those with HCC: hypoenhancement in the PVP images (88.9% vs. 64.4%),unclear boundary in the intratumoral nonenhanced area (71.1% vs. 37.8%), and partial washout (71.1% vs. 40.0%).The baseline CEUS imaging features of all lesions are presented in Table 2.

The efficiency of the imaging models for cHCC-CCA The CEUS-predominant model
The CEUS-predominant model was developed by combining CEUS features and tumor markers (AFP > 400  4).The AUC value of the prediction model was 0.720 (95% CI: 0.632, 0.808).The sensitivity, specificity, and accuracy were 55.6%, 80.0%, and 71.9%, respectively.Regression coefficient-based nomograms were constructed based on the CEUS-predominant model (Fig. 3a).The calibration curve of the nomogram for the probability of cHCC-CCA demonstrated good agreement between prediction and observation (eFigure 1a).The H-L test yielded a nonsignificant statistic (p = 1.000).

The CECT-predominant model
The

Discussion
Combining tumor biomarkers and imaging features is critical in diagnosing cHCC-CCA due to its overlapping features with HCC.This propensity score-matched study found that approximately 44.4% of cHCC-CCAs on CEUS and 26.7% of cHCC-CCAs on CECT were evaluated as LR-5, which can easily mimic HCC.Therefore, we constructed and compared two imaging-predominant diagnostic models based on clinical data and nodule features on CEUS and CECT imaging to identify cHCC-CCA.The results indicated that the CECTpredominant model exhibited nearly perfect diagnostic sensitivity (93.3%), which was significantly higher than that of the CEUS-predominant model (55.6%; p < 0.001).On the other hand, the CEUS-predominant model demonstrated commendable diagnostic specificity, particularly for lesions smaller than 5 cm (92.0% vs. 70.0%;p = 0.013).
Cirrhosis detected by CECT is highly suggestive of HCC.In this study, we found that few at-risk patients with cHCC-CCA had a cirrhotic liver background due to the different origins of HCC and cHCC-CCA, which is similar to the findings of the latest studies [24,[29][30][31].Additionally, this study revealed that cHCC-CCA patients exhibited higher rates of nonrim APHE with a volume < 50%.This finding is congruent with a previous study that reported a larger HCC component (p = 0.014) and a smaller ICC component (p = 0.001) in the hypervascular group of cHCC-CCA patients during pathological analysis [32].In addition, capsular enhancement is usually considered a major imaging feature of HCC [24,33].In the present study, we observed a higher frequency of incomplete tumor capsules in cHCC-CCA than in HCCs.Similar to the observation of "unclear boundaries" on CEUS, the presence of an incomplete capsule is likely associated with infiltrative tumor growth of the ICC portion in cHCC-CCA [15,34].Interestingly, ten out of 12 cHCC-CCAs, classified as LR-5 based on CECT LI-RADS classification, were indeed reclassified as cHCC-CCA based on the CECT-predominant model in this study, which may help improve the diagnostic specificity of LR-5 for HCC in future clinical practice.Based on these features, the likelihood of this mass being diagnosed as cHCC-CCA was smaller than 30% according to the CEUS-predominant model (B).There was no obvious cirrhotic liver background, and the mass showed low density on abdominal CT image (A, d), rim enhancement and < 50% nonrim enhancement (mainly the right posterior part of the lesion, arrow) in the arterial phase (A, e), "washout" absence, nonsmooth tumor margin, and a thin incomplete enhanced capsule (arrow) seen in the portal venous phase (A, f ).Based on these features, the likelihood of this mass being diagnosed as cHCC-CCA was higher than 90.0% according to the CECT-predominant model (C).The mass was pathologically proven to be combined hepatocellular-cholangiocarcinoma (See figure on next page.)Fig. 4 (See legend on previous page.) On CEUS images, the presence of unclear boundaries in the intratumoral nonenhanced areas was an independent risk factor for cHCC-CCA.This might be elucidated by the fibrotic pathological findings (relying on the ICC component), similar to previous findings [21,35].The presence of washout on CEUS with SonoVue reflects the intratumoral vascular supply.Therefore, ICCs often present earlier and with more complete washout compared to HCCs [36].Consequently, partial washout could frequently be observed in cHCC-CCA lesions that contain both HCC and ICC components.Notably, tumor differentiation is correlated with the presence of washout, as demonstrated by the findings of Iavarone et al. [37].This study also observed that grade 2-3 HCCs exhibited higher rates of partial washout than lower grade HCCs (grade 1 or 2) (19/37 vs. 17/52; p = 0.079).This finding might provide valuable prognostic information for future studies in this field.
Several studies have evaluated the performance of imaging characteristics in differentiating cHCC-CCA and HCC in recent years [15,18,31,38,39].The model developed in this study, which combined CECT features and CA 19-9 levels to distinguish cHCC-CCA, was put into a visual form as a nomogram and demonstrated a remarkable sensitivity of 93.3%, showing better performance than previous ones (ranging from 40 to 73.8% [15,18,38].We hope that our findings can offer valuable guidance in two aspects.First, the remarkably high sensitivity for cHCC-CCA could effectively diagnose lesions that do not support a definite diagnosis of HCC, which might improve the diagnostic specificity of HCC in routine clinical practice.In addition, the CECT-predominant model exhibited a low specificity of 63.3%, which could lead to misdiagnosis or underdiagnosis of a higher number of HCC cases, which would limit the therapeutic options for HCC (e.g., liver transplantation).Given this, adding CEUS to CECT could improve the overall diagnostic accuracy, especially for lesions less than 5 cm.Some limitations of this study should be mentioned.First, there was an unavoidable selection bias due to the single-center retrospective nature of the study, although we used PSM to lessen this bias.Second, no validation data were available to test and refine our models due to the limited size of the cHCC-CCA population.Third, we did not include ICC patients in this differential diagnostic study due to the limited number of ICC patients with HCC risk factors.Finally, the results of this study were based on a case-control design rather than a cohort design, which might not reflect real-world clinical epidemiological conditions.Therefore, large-scale multicenter studies are warranted to validate our findings.

Conclusions
The CECT-predominant model provides higher diagnostic sensitivity compared to the CEUS-predominant model for cHCC-CCA.Combining the CECT features with serum CA 19-9 > 100 U/mL showed excellent diagnostic sensitivity in differentiating cHCC-CCA from HCC, while the CEUS features could enhance diagnostic specificity, especially in the ≤ 5 cm subgroup.

Fig. 2
Fig. 2 The definition of the partial imaging features of the lesions on CEUS (a) and CECT (b)

Fig. 4
Fig.4CEUS and CECT images of a 56-year-old man with chronic hepatitis B and CA 19-9 < 100 U/mL.A 7.3-cm mass was detected in segment IV of the liver (A).A hypoechoic mass with poor boundary on conventional ultrasound (A, a); on CEUS, the mass showed hyperenhancement, a nonsmooth tumor margin (stars), and tumor supply artery (arrowhead) at 18 s (A, b); in the late phase (179 s), the hyperenhanced area in the arterial phase of mass exhibited partial washout with partial isoenhancement (stars) and partial hypoenhancement area (arrowhead, A, c).Based on these features, the likelihood of this mass being diagnosed as cHCC-CCA was smaller than 30% according to the CEUS-predominant model (B).There was no obvious cirrhotic liver background, and the mass showed low density on abdominal CT image (A, d), rim enhancement and < 50% nonrim enhancement (mainly the right posterior part of the lesion, arrow) in the arterial phase (A, e), "washout" absence, nonsmooth tumor margin, and a thin incomplete enhanced capsule (arrow) seen in the portal venous phase (A, f ).Based on these features, the likelihood of this mass being diagnosed as cHCC-CCA was higher than 90.0% according to the CECT-predominant model (C).The mass was pathologically proven to be combined hepatocellular-cholangiocarcinoma

Fig. 5
Fig.5 The diagnostic performance the CEUS-predominant and CECT-predominant models was assessed through ROC curve and AUC analyses

Table 1
Basic clinical and pathological characteristics of patients with cHCC-CCA and HCCcHCC-CCA combined hepatocellular-cholangiocarcinoma, HCC hepatocellular carcinoma, HBV hepatitis B virus, HCV hepatitis C virus, AFP alpha-fetoprotein, CA 19-9 carbohydrate antigen 19-9 # Clinical variables were subjected to further logistic regression analysis when p values were < 0.05 in the χ 2 test or Fisher's exact test, as appropriate a Unless stated otherwise, data in parentheses are counts (percentages) b For the multiple nodules in the liver, only the largest nodule was analyzed in this study c The components of HCC were included in both grades 2 and 3 based on pathological results

Table 2
The CEUS features of included lesions cHCC-CCA combined hepatocellular-cholangiocarcinoma, HCC hepatocellular carcinoma, CEUS contrast-enhanced ultrasound, AP arterial phase, PVP portal venous phase, LP late phase, NA not available, APHE arterial phase hyperenhancement, LI-RADS liver imaging reporting and data system, LR liver imaging reporting and data system category # Categorical variables were compared by the χ 2 test or Fisher's exact test a Unless otherwise indicated, data are number of patients, with percentage in parentheses

Table 3
The CECT features of included lesions cHCC-CCA combined hepatocellular-cholangiocarcinoma, HCC hepatocellular carcinoma, CECT contrast-enhanced computed tomography, APHE arterial phase hyperenhancement, LI-RADS liver imaging reporting and data system, LR liver imaging reporting and data system category # Categorical variables were compared by the χ 2 test or Fisher's exact test a Unless otherwise indicated, data are number of patients, with percentage in parentheses b Lesion with LR-M features means that lesions have LR-M category features with or without tumor thrombus in a vessel on CECT

Table 4
Univariate and multivariate logistic regression analyses for diagnosing cHCC-CCA with the CEUS-predominant and CECTpredominant models cHCC-CCA combined hepatocellular-cholangiocarcinoma, CEUS contrast-enhanced ultrasound, CECT contrast-enhanced computed tomography, CA 19-9 carbohydrate antigen 19-9, OR odds ratio, PVP portal venous phase, APHE arterial phase hyperenhancement, CI confidence interval, LR liver imaging reporting and data system category a Collinearity exists among these variables (the same words on the right upper), variables with higher odds ratio were used by multivariate regression analysis b Tumor growth subtype: type 2 (single nodule with extranodular growth); type 3 (multiple confluent nodules) c Lesion with LR-M features means that lesions have LR-M category feature accompanying with/without tumor thrombus in vascular on CECT

Table 5
Comparison of the diagnostic performance between the CEUS-predominant and CECT-predominant modelsCEUS contrast-enhanced ultrasound, CECT contrast-enhanced computed tomography, AUC area under the curve, CI confidence interval a The discrimination of AUC value was considered fair (AUC < 0.6), moderate (AUC, 0.6-0.75),and substantial (AUC > 0.75), respectively # p value was obtained from the comparison between the CEUS-predominant and CECT-predominant models by the McNemar test.p < 0.05 was considered to indicate a statistical difference