Abstract
Objective
Liver Imaging Reporting and Data System category M (LR-M) is associated with a poor prognosis in intrahepatic cholangiocarcinoma (iCCA). However, the prognostic value of individual LR-M features across imaging modalities remains unclear. We aimed to determine, through intraindividual comparisons, whether the prognostic value of individual LR-M features differs between CT and MRI in patients with iCCA at risk for hepatocellular carcinoma.
Materials and Methods
This retrospective study included consecutive patients with cirrhosis or chronic hepatitis B who underwent curative-intent resection for iCCA and preoperative multiphasic CT and MRI between 2010 and 2019. CT and MRI images were reviewed in separate sessions ≥4 weeks apart. Cox proportional hazards regression analyses with and without adjustment for LI-RADS categorization (non-LR-M vs. single-modality LR-M or dual-modality LR-M) were performed to evaluate the associations of CT- and MRI-based targetoid and non-targetoid LR-M features and other variables with recurrence-free survival (RFS) and overall survival (OS).
Results
A cohort of 122 patients (mean age, 57.9 ± 10.4 years) was included. MRI demonstrated significantly higher frequencies than CT for targetoid (84.4% vs. 73.8%, P = 0.007) and non-targetoid features (43.4% vs. 20.5%, P < 0.001), resulting in a higher rate of LR-M categorization (90.2% vs. 81.1%, P = 0.003). In the multivariable analysis, CT-based targetoid features (hazard ratio [HR] = 2.31 for RFS, P = 0.010; HR = 1.85 for OS, P = 0.043) and MRI-based non-targetoid features (HR = 2.15 for RFS, P = 0.005; HR = 2.09 for OS, P = 0.008) were independently associated with poorer survival. After adjustment for LI-RADS categorization, MRI-based non-targetoid features remained independently associated with poorer RFS (HR = 1.76, P = 0.041) and OS (HR = 1.79, P = 0.038). Dual-modality LR-M designation was independently associated with poorer RFS (HR = 8.02, P = 0.042).
Conclusion
The prognostic implications of LR-M features in iCCA vary by imaging modality and feature type. CT-based targetoid and MRI-based non-targetoid features were associated with poorer survival, and the associations between MRI-based non-targetoid features and survival remained significant after adjustment for LI-RADS categorization.
Keywords: Liver neoplasms, Cholangiocarcinoma, Computed tomography, Magnetic resonance imaging, Liver imaging reporting and data system, Prognosis
INTRODUCTION
Intrahepatic cholangiocarcinoma (iCCA) is the second-most common primary hepatic malignancy, and its global incidence continues to increase in parallel with the increasing prevalence of risk factors such as type 2 diabetes mellitus and cirrhosis [1,2]. Despite the advances in our understanding of iCCA tumor biology, surgical resection remains the only curative treatment option. Nevertheless, postoperative recurrence occurs in approximately 20%–50% of cases within 12–24 months, and fewer than 40% of patients survive beyond 5 years, even after curative-intent resection [3,4,5]. Therefore, identifying reliable biomarkers for prognostic stratification of these cancers before surgery is of clinical importance.
The Liver Imaging Reporting and Data System (LI-RADS) provides a standardized framework for characterization and categorization of liver observations on computed tomography (CT) and magnetic resonance imaging (MRI) in patients at risk of hepatocellular carcinoma (HCC). Among the LI-RADS categories, category M (LR-M) is assigned to observations of probable or definite malignancy, but is not specific to HCC and is commonly associated with non-HCC malignancies, including iCCA [6,7]. Beyond its diagnostic role, emerging evidence suggests that the LI-RADS categorization system may also provide prognostic information. Recent studies have shown that iCCAs categorized as LR-M are associated with poorer postsurgical outcomes than those categorized as LR-4/5 [8,9,10]. However, although LI-RADS categories and imaging features have been reported to differ substantially between CT and MRI [11,12,13,14], most previous studies evaluating the prognostic value of LR-M have primarily focused on MRI. Consequently, the prognostic implications of CT-based LR-M classifications remain poorly understood.
Notably, the LR-M category encompasses a heterogeneous spectrum of imaging characteristics [6]. Although both targetoid and non-targetoid features are included within this category, they play different roles in the LI-RADS diagnostic algorithm. Specifically, the presence of any targetoid feature, such as rim arterial-phase hyperenhancement (APHE), peripheral washout, or delayed central enhancement, directly leads to an LR-M categorization. In contrast, non-targetoid features, including an infiltrative appearance, marked diffusion restriction, and necrosis or severe ischemia, will only result in an LR-M designation if the observation does not meet the LR-5 criteria. Given these differences, the biological behavior and clinical implications of LR-M observations may vary in accordance with both the imaging modality and feature type. However, most previous studies evaluating the prognostic value of LR-M in iCCA treated this classification as a single category without distinguishing between these feature types or considering potential modality-specific effects.
Therefore, we aimed to determine, through intraindividual comparisons in patients with surgically resected iCCA who were at risk for HCC, whether the prognostic value of individual LR-M features differs between CT and MRI.
MATERIALS AND METHODS
This retrospective study was approved by the Institutional Review Board of Asan Medical Center (IRB No. 2025-1589), and the requirement for written informed consent was waived.
Study Participants
The electronic medical records of the study institution were reviewed to identify consecutive patients who underwent hepatic resection for treatment-naïve Icca between January 2010 and June 2019. The inclusion criteria were as follows: 1) age ≥18 years and 2) at risk for HCC according to LI-RADS (i.e., presence of cirrhosis or chronic hepatitis B virus infection). Patients were excluded sequentially from the study cohort for the following reasons: 1) absence of preoperative multiphasic CT or MRI performed within 3 months before surgery, 2) presence of concurrent HCC or other primary malignancy, 3) hepatic resection performed with palliative rather than curative intent, 4) Icca arising from intraductal papillary neoplasm of the bile duct or mucinous cystic neoplasm, 5) non-visualization of the tumor on CT or MRI, 6) macrovascular invasion demonstrated on CT and MRI, and 7) severe image artifacts. In accordance with clinical practice guidelines [15,16], patients with suspected or confirmed Icca routinely underwent preoperative multiphasic CT and MRI for staging evaluation. A flow diagram summarizing patient inclusion and exclusion criteria is shown in Figure 1.
Fig. 1. Flow diagram for selection of the study population. iCCA = intrahepatic cholangiocarcinoma, HCC = hepatocellular carcinoma, IPNB = intraductal papillary neoplasm of the bile duct, MCN = mucinous cystic neoplasm.
Collection of Clinical Data and Pathologic Analysis
Clinical data were extracted from the electronic medical records of the patients included in the study cohort, including information regarding age, sex, viral serology, serum carbohydrate antigen 19-9 (CA 19-9) level measured within 1 month before surgery, extent of hepatic resection, and adjuvant treatments.
The diagnosis of iCCA was confirmed on the basis of the pathological examination of resected specimens in accordance with the 2019 World Health Organization classification [17]. These pathological assessments included evaluation of tumor size, macroscopic growth pattern (mass-forming, periductal infiltrating, or intraductal growth), histological differentiation grade, presence of microvascular or macrovascular invasion, perineural invasion, resection margin status, lymph node metastasis, and background liver cirrhosis. Tumor staging was performed in accordance with the 8th edition of the American Joint Committee on Cancer TNM staging system [18]. Detailed staging criteria are provided in Supplementary Table 1.
CT and MRI Acquisition
All study participants underwent both multiphasic CT and MRI examinations in accordance with the technical standards of LI-RADS v2018 [6]. Contrast-enhanced MRI was performed using hepatobiliary contrast agents, including gadoxetic acid (Eovist; Bayer Healthcare, Berlin, Germany; n = 98) and gadobenate dimeglumine (MultiHance; Bracco Diagnostics, Milan, Italy; n = 7), or the extracellular contrast agent gadoterate meglumine (Dotarem; Guerbet, Villepinte, France; n = 17). Detailed descriptions of the CT and MRI acquisition techniques and imaging parameters used are provided in the Supplementary Methods and in Supplementary Table 2.
Image Assessments
An investigator with 16 years of experience in hepatic imaging identified and annotated the target iCCA on both CT and MRI examinations on the basis of the pathological report to ensure accurate lesion localization. This analysis was independent of image interpretation. The investigator measured tumor size as the greatest axial diameter of the annotated lesion. In patients with multiple tumors, the largest tumor was selected for annotation.
Two board-certified abdominal radiologists with 6 and 14 years of experience in hepatic imaging independently reviewed the examinations in random order while remaining blinded to all clinicopathological data and patient outcomes. Before image interpretation, the readers reviewed and standardized the definitions of imaging features using representative cases that were not included in the study. Image interpretation was conducted in two separate sessions, with CT images reviewed first and MRI images reviewed at least four weeks later to minimize recall bias. For each tumor, the readers recorded the presence of major feature, ancillary features, and LR-M features in accordance with the LI-RADS lexicon and LI-RADS v2018 diagnostic algorithm [6]. In accordance with LI-RADS, non-peripheral washout on MRI was assessed on the portal venous phase in examinations performed with hepatobiliary contrast agents, whereas both the portal venous and delayed phases were evaluated in examinations performed with extracellular contrast agents.
The LR-M features were categorized as targetoid or non-targetoid. Targetoid features included rim APHE, peripheral washout, delayed central enhancement, targetoid diffusion restriction, and targetoid transitional phase or hepatobiliary-phase (HBP) appearance. Targetoid transitional phase or HBP appearance was assessed only during examinations performed using hepatobiliary contrast agents. Non-targetoid features included infiltrative appearance, marked diffusion restriction, and necrosis or severe ischemia. On the basis of the recorded features, each tumor was assigned a final LI-RADS category (LR-4, LR-5 and LR-M), and CT and MRI data were evaluated separately. In cases involving disagreement between the two primary readers regarding the presence or absence of imaging features, a consensus was reached through a joint review with a third abdominal radiologist with 20 years of experience in hepatic imaging. Definitions of all imaging features are detailed in Supplementary Table 3.
Follow-Up and Outcome Measures
After surgical resection, patients with iCCA underwent follow-up at 3–6-month intervals using contrast-enhanced CT or MRI and serum tumor marker assessments. The study endpoints were recurrence-free survival (RFS) and overall survival (OS). RFS was defined as the interval from surgery to the first documented intrahepatic or extrahepatic recurrence or death, whichever occurred first. OS was defined as the interval from surgery to death from any cause. Patients without events were censored on the date of their last follow-up.
Statistical Analysis
Continuous variables were expressed as the mean ± standard deviation or as the median with interquartile range (IQR), depending on the data distribution assessed by normality testing. Categorical variables were presented as numbers and percentages. Interobserver and inter-modality agreement for imaging features was evaluated using Cohen’s κ statistic.
The proportions of LR-M and LR-5 classifications on CT and MRI were compared using McNemar tests. On the basis of the combined CT and MRI LI-RADS categorization, patients with iCCA were classified into three groups: non-LR-M (LR-4/5 on both CT and MRI), single-modality LR-M (LR-M on either CT or MRI), and dual-modality LR-M (LR-M on both CT and MRI). The paired proportions of patients demonstrating individual targetoid and non-targetoid features, as well as the presence of at least one targetoid or non-targetoid feature on CT and MRI, were also compared using the McNemar test.
RFS and OS values were estimated using the Kaplan–Meier method, and groups were compared using the log-rank test. For survival analyses, the presence of at least one targetoid or non-targetoid feature on CT and MRI was treated as a binary variable. An additional exploratory analysis was performed using a binary variable defined by the presence of at least one MRI-based targetoid feature assessable on both modalities (i.e., rim APHE, peripheral washout, or delayed central enhancement). Cox proportional hazards regression analyses were used to identify variables associated with RFS and OS, and the results were reported as hazard ratios (HRs) and 95% confidence intervals (CIs). Variables with a P-value <0.05 in univariable analysis were entered into multivariable models. Multicollinearity was evaluated using Pearson correlation coefficients and variance inflation factors. Additional exploratory univariable Cox regression analyses were performed to evaluate the associations between individual LR-M features and survival outcomes.
All statistical analyses were performed using SPSS (version 29.0; IBM Corp., Armonk, NY, USA) and R (version 4.3.1; R Foundation for Statistical Computing, Vienna, Austria). Statistical significance was defined as a two-sided P-value <0.05.
RESULTS
Study Patient Characteristics
Among 226 eligible patients with iCCA initially screened in our institutional database, 104 were sequentially excluded from further analysis due to unavailability of CT or MRI results within 3 months before surgery (n = 31), the presence of concurrent HCC or other primary malignancy (n = 13), palliative hepatic resection (n = 35), iCCA arising from an intraductal papillary neoplasm of the bile duct or mucinous cystic neoplasm (n = 10), nonvisualization of the tumor (n = 4), macrovascular invasion on CT and MRI (n = 6), or severe MRI artifacts (n = 5). The final study cohort consisted of 122 patients (mean age, 57.9 ± 10.4 years; 77 [63.1%] male). Among these patients, 44.3% (54/122) had histologically confirmed cirrhosis, and 55.7% (68/122) had chronic hepatitis B without cirrhosis (Table 1). The mean pathologic tumor size was 4.6 ± 2.5 cm. Most tumors showed a purely mass-forming growth pattern (95.9% [117/122]), were well or moderately differentiated (80.5% [95/118]), and were classified as TNM stage I or II (74.5% [91/122]).
Table 1. Clinical and pathologic characteristics of the 122 study participants.
| Characteristic | Value | ||
|---|---|---|---|
| Clinical characteristics | |||
| Age, yrs | 57.9 ± 10.4 | ||
| Sex | |||
| Male | 77 (63.1) | ||
| Female | 45 (36.9) | ||
| Risk factor | |||
| Cirrhosis | 54 (44.3) | ||
| Chronic hepatitis B without cirrhosis | 68 (55.7) | ||
| Serum CA 19-9, U/mL | 19.2 (9.8–74.1) | ||
| Resection procedure | |||
| Major hepatectomy | 52 (42.6) | ||
| Minor hepatectomy | 70 (57.4) | ||
| Received adjuvant chemotherapy | 52 (42.6) | ||
| Pathologic characteristics | |||
| Tumor size, cm | 4.6 ± 2.5 | ||
| Macroscopic features | |||
| MF type, purely | 117 (95.9) | ||
| MF combined with PI or IG type | 5 (4.1) | ||
| Differentiation* | |||
| Well or moderate | 95 (80.5) | ||
| Poor | 23 (19.5) | ||
| Vascular invasion | 59 (48.4) | ||
| Perineural invasion | 34 (27.9) | ||
| Positive resection margin | 9 (7.4) | ||
| TNM stage, AJCC 8th edition | |||
| I | 58 (47.5) | ||
| II | 33 (27.0) | ||
| III | 26 (21.3) | ||
| IV | 5 (4.1) | ||
Values are expressed as mean ± standard deviation, number (%) of patients, or median (interquartile range).
*Data were unavailable for four patients.
CA = carbohydrate antigen, MF = mass-forming, PI = periductal infiltrative, IG = intraductal growing, AJCC = American Joint Committee on Cancer
Distribution of LI-RADS Categories on CT and MRI
The mean interval between CT and MRI examinations was 14 ± 14 days (range, 0–63 days). Among the 122 observations in the study cohort, 9 (7.4%), 14 (11.5%), and 99 (81.1%) lesions were categorized as LR-4, LR-5, and LR-M on CT, respectively, whereas 8 (6.6%), 4 (3.3%), and 110 (90.2%) lesions were categorized as LR-4, LR-5, and LR-M on MRI, respectively (Supplementary Fig. 1). The proportion of LR-M designations was significantly higher on MRI than on CT (90.2% vs. 81.1%, P = 0.003). Conversely, CT demonstrated a significantly higher proportion of LR-5 designations than MRI (11.5% vs. 3.3%; P = 0.006). Among the 13 patients with discordant LR-M categorization, 12 were classified as having MRI-only LR-M and one as having CT-only LR-M. Based on the combined CT and MRI LI-RADS assessments, 98 patients (80.3%) were classified into the dual-modality LR-M group, 13 (10.7%) into the single-modality LR-M group, and 11 (9.0%) into the non-LR-M group.
Distribution of LR-M Features on CT and MRI
The frequency of tumors demonstrating at least one targetoid feature was significantly higher on MRI than on CT (84.4% [103/122] vs. 73.8% [90/122], P = 0.007). In addition, MRI demonstrated a significantly higher frequency of tumors with at least one non-targetoid feature than did CT (43.4% [53/122] vs. 20.5% [25/122], P < 0.001; Table 2). Among the observed targetoid features, delayed central enhancement was the most frequent finding on both CT (65.6% [80/122]) and MRI (69.7% [85/122]). Among the detected non-targetoid features, an infiltrative appearance was the most common feature on CT (18.0% [22/122]), whereas marked diffusion restriction was the most common feature on MRI (25.4% [31/122]).
Table 2. CT and MRI features of the 122 intrahepatic cholangiocarcinomas in the study cohort.
| Imaging features | CT | MRI | P | |
|---|---|---|---|---|
| LI-RADS category | ||||
| LR-4 | 9 (7.4) | 8 (6.6) | >0.999 | |
| LR-5 | 14 (11.5) | 4 (3.3) | 0.006 | |
| LR-M | 99 (81.1) | 110 (90.2) | 0.003 | |
| Targetoid features | ||||
| Rim APHE | 73 (59.8) | 68 (55.7) | 0.533 | |
| Peripheral washout | 20 (16.4) | 16 (13.1) | 0.585 | |
| Delayed central enhancement | 80 (65.6) | 85 (69.7) | 0.473 | |
| Targetoid restriction | N/A | 44 (36.1) | N/A | |
| Targetoid TP or HBP appearance | N/A | 32 (26.2) | N/A | |
| At least one targetoid feature | 90 (73.8) | 103 (84.4) | 0.007 | |
| Non-targetoid features | ||||
| Infiltrative appearance | 22 (18.0) | 20 (16.4) | 0.824 | |
| Marked diffusion restriction | N/A | 31 (25.4) | N/A | |
| Necrosis or severe ischemia | 8 (6.6) | 9 (7.4) | >0.999 | |
| At least one non-targetoid feature | 25 (20.5) | 53 (43.4) | <0.001 | |
Values are expressed as the number (%) of patients.
LI-RADS = Liver Imaging Reporting and Data System, LR-4 = LI-RADS category 4, LR-5 = LI-RADS category 5, LR-M = LI-RADS category M, APHE = arterial phase hyperenhancement, N/A = not applicable, TP = transitional phase, HBP = hepatobiliary phase
The inter-modality agreement between CT and MRI for the LR-M imaging features was generally fair to moderate (Supplementary Table 4). For features assessable on both modalities, agreement for the presence of at least one targetoid feature was moderate (κ = 0.42), whereas agreement for the presence of at least one non-targetoid feature was fair (κ = 0.22).
The interobserver agreement for targetoid and non-targetoid features was moderate-to-substantial on both CT and MRI (Supplementary Table 5). Among the LR-M features, rim APHE showed the highest kappa value on CT (κ = 0.65), whereas infiltrative appearance demonstrated the highest kappa value on MRI (κ = 0.71).
Univariable Analysis of Survival in Relation to LI-RADS Concordance and LR-M Features
Over a median follow-up period of 51.1 months (IQR, 21.6–109.4 months), tumor recurrence and death occurred in 73 (59.8%) and 75 (61.5%) of the 122 study patients, respectively. Among the 75 patients who died, 6 had no tumor recurrence during follow-up. Among patients who experienced recurrence, 22 (30.1%) had intrahepatic recurrence only, 40 (54.8%) had extrahepatic recurrence only, and 11 (15.1%) had both intrahepatic and extrahepatic disease. When stratified on the basis of LI-RADS concordance across imaging modalities, the dual-modality LR-M group demonstrated significantly poorer RFS and OS than the single-modality LR-M and non-LR-M groups (all P < 0.05; Supplementary Fig. 2).
For targetoid LR-M features, patients with at least one CT-based targetoid feature had significantly poorer RFS (HR, 2.56; 95% CI, 1.37–4.75; P = 0.003; Fig. 2A) and OS (HR, 2.05; 95% CI, 1.15–3.67; P = 0.015; Fig. 2B) than those without such features. In contrast, the presence of at least one MRI-based targetoid feature showed trends toward shorter RFS (HR, 1.58; 95% CI, 0.78–3.17; P = 0.203; Fig. 2C) and OS (HR, 1.52; 95% CI, 0.76–3.06; P = 0.238; Fig. 2D), although these associations did not reach statistical significance. Similarly, when the analysis was restricted to targetoid features assessable on both CT and MRI, the presence of at least one MRI-based targetoid feature continued to show trends toward shorter RFS (HR, 1.75; 95% CI, 0.87–3.51; P = 0.118; Supplementary Fig. 3A) and OS (HR, 1.88; 95% CI, 0.96–3.66; P = 0.064; Supplementary Fig. 3B), although the associations remained statistically nonsignificant.
Fig. 2. Kaplan–Meier curves demonstrating the RFS and OS outcomes in the study population in relation to the presence of at least one targetoid feature on CT (A, B) and MRI (C, D). Patients with at least one CT-based targetoid feature had significantly shorter RFS (A) and OS (B) than those without such features (P = 0.003 and P = 0.015, respectively). In contrast, MRI-based targetoid features showed trends toward shorter RFS (C) and OS (D), although neither association reached statistical significance (P = 0.203 and P = 0.238, respectively). RFS = recurrence-free survival, OS = overall survival, HR = hazard ratio, CI = confidence interval.
For non-targetoid LR-M features, the presence of at least one CT-based non-targetoid feature showed trends toward shorter RFS (HR, 1.65; 95% CI, 0.95–2.84; P = 0.074; Fig. 3A) and OS (HR, 1.42; 95% CI, 0.84–2.42; P = 0.195; Fig. 3B), although the differences were not statistically significant. In contrast, patients with at least one MRI-based non-targetoid feature had significantly poorer RFS (HR, 2.84; 95% CI, 1.78–4.55; P < 0.001; Fig. 3C) and OS (HR, 2.80; 95% CI, 1.76–4.46; P < 0.001; Fig. 3D) than those without such features. Detailed survival data, including median survival times and 1-, 5-, and 10-year survival rates, are provided in Supplementary Table 6.
Fig. 3. Kaplan–Meier curves demonstrating RFS and OS in the study population in relation to the presence of at least one non-targetoid feature on CT (A, B) and MRI (C, D). The presence of CT-based non-targetoid features showed trends toward shorter RFS (A) and OS (B), although neither association reached statistical significance (P = 0.074 and P = 0.195, respectively). In contrast, patients with at least one MRI-based non-targetoid feature had significantly shorter RFS (C) and OS (D) values than those without such features (both P < 0.001). RFS = recurrence-free survival, OS = overall survival, HR = hazard ratio, CI = confidence interval.
Multivariable Analysis of the Association of LR-M Features With Survival
Multivariable Cox regression analysis was performed to identify the factors independently associated with RFS and OS. Because LI-RADS categorization and the presence of at least one CT-based targetoid feature were highly correlated (Pearson correlation coefficient, 0.78), whereas all other significant variables demonstrated correlation coefficients of 0.35 or less, separate multivariable models were constructed to minimize the effects of multicollinearity. Model 1 included CT-based targetoid features, but excluded LI-RADS categorization, whereas Model 2 included LI-RADS categorization, but excluded CT-based targetoid features (Tables 3, 4).
Table 3. Univariable and multivariable Cox regression analyses of the factors associated with recurrence-free survival.
| Variables | Univariable analysis | Multivariable analysis: model 1* | Multivariable analysis: model 2† | ||||
|---|---|---|---|---|---|---|---|
| HR (95% CI) | P | HR (95% CI) | P | HR (95% CI) | P | ||
| Age (≥60 years) | 0.85 (0.54−1.35) | 0.490 | |||||
| Male sex | 1.83 (1.10−3.04) | 0.020 | 1.77 (1.04−3.00) | 0.034 | 1.91 (1.13−3.24) | 0.016 | |
| Log transformed serum CA 19-9, U/mL | 1.29 (1.12−1.48) | <0.001 | 1.08 (0.92−1.26) | 0.341 | 1.12 (0.96−1.29) | 0.147 | |
| Extent of hepatectomy (major hepatectomy) | 1.33 (0.84−2.10) | 0.231 | |||||
| Cirrhosis by B viral hepatitis infection | 0.87 (0.55−1.39) | 0.564 | |||||
| Macroscopic feature (mass-forming type, purely) | 0.58 (0.21−1.59) | 0.290 | |||||
| Tumor differentiation (poorly differentiated) | 1.53 (0.88−2.67) | 0.136 | |||||
| Perineural invasion | 3.31 (2.04−5.37) | <0.001 | 2.30 (1.31−4.04) | 0.004 | 2.02 (1.13−3.61) | 0.018 | |
| AJCC TNM stage (III or IV) | 4.79 (2.91−7.89) | <0.001 | 2.68 (1.48−4.85) | 0.001 | 2.51 (1.38−4.58) | 0.003 | |
| LI-RADS categorization | |||||||
| Non-LR-M | Reference | Reference | |||||
| Single-modality LR-M | 4.90 (0.57−41.92) | 0.147 | 4.02 (0.46−34.89) | 0.208 | |||
| Dual-modality LR-M | 13.94 (1.93−100.65) | 0.009 | 8.02 (1.08−59.43) | 0.042 | |||
| At least one CT-based targetoid feature | 2.56 (1.37−4.75) | 0.003 | 2.31 (1.22−4.38) | 0.010 | |||
| At least one MRI-based targetoid feature | 1.58 (0.78−3.17) | 0.203 | |||||
| At least one CT-based non-targetoid feature | 1.65 (0.95−2.84) | 0.074 | |||||
| At least one MRI-based non-targetoid feature | 2.84 (1.78−4.55) | <0.001 | 2.15 (1.26−3.68) | 0.005 | 1.76 (1.02−3.02) | 0.041 | |
*Model 1 included at least one CT-based targetoid feature but excluded LI-RADS categorization because of multicollinearity, †Model 2 included LI-RADS categorization but excluded at least one CT-based targetoid feature because of multicollinearity.
LI-RADS = Liver Imaging Reporting and Data System, HR = hazard ratio, CI = confidence interval, CA = carbohydrate antigen, AJCC = American Joint Committee on Cancer, LR-M = LI-RADS category M
Table 4. Univariable and multivariable Cox regression analyses of the factors associated with overall survival.
| Variables | Univariable analysis | Multivariable analysis: model 1* | Multivariable analysis: model 2† | ||||
|---|---|---|---|---|---|---|---|
| HR (95% CI) | P | HR (95% CI) | P | HR (95% CI) | P | ||
| Age (≥60 years) | 0.98 (0.62−1.55) | 0.942 | |||||
| Male sex | 1.84 (1.11−3.05) | 0.018 | 2.14 (1.25−3.68) | 0.006 | 2.15 (1.25−3.69) | 0.006 | |
| Log transformed serum CA 19-9, U/mL | 1.41 (1.23−1.62) | <0.001 | 1.24 (1.07−1.44) | 0.005 | 1.25 (1.08−1.44) | 0.003 | |
| Extent of hepatectomy (major hepatectomy) | 1.06 (0.67−1.68) | 0.796 | |||||
| Cirrhosis by B viral hepatitis infection | 1.11 (0.71−1.75) | 0.649 | |||||
| Macroscopic feature (mass-forming type, purely) | 0.71 (0.26−1.96) | 0.510 | |||||
| Tumor differentiation (poorly differentiated) | 1.22 (0.69−2.15) | 0.496 | |||||
| Perineural invasion | 3.19 (1.97−5.15) | <0.001 | 2.60 (1.47−4.58) | <0.001 | 2.24 (1.26−3.98) | 0.006 | |
| AJCC TNM stage (III or IV) | 4.75 (2.91−7.77) | <0.001 | 2.47 (1.38−4.42) | 0.002 | 2.57 (1.43−4.62) | 0.002 | |
| LI-RADS categorization | |||||||
| Non-LR-M | Reference | Reference | |||||
| Single-modality LR-M | 1.90 (0.35−10.37) | 0.459 | 1.40 (0.25−7.83) | 0.700 | |||
| Dual-modality LR-M | 6.65 (1.63−27.17) | 0.008 | 3.48 (0.82−14.74) | 0.091 | |||
| At least one CT-based targetoid feature | 2.05 (1.15−3.67) | 0.015 | 1.85 (1.02−3.36) | 0.043 | |||
| At least one MRI-based targetoid feature | 1.52 (0.76−3.06) | 0.238 | |||||
| At least one CT-based non-targetoid feature | 1.42 (0.84−2.42) | 0.195 | |||||
| At least one MRI-based non-targetoid feature | 2.80 (1.76−4.46) | <0.001 | 2.09 (1.21−3.61) | 0.008 | 1.79 (1.03−3.12) | 0.038 | |
*Model 1 included at least one CT-based targetoid feature but excluded LI-RADS categorization because of multicollinearity, †Model 2 included LI-RADS categorization but excluded at least one CT-based targetoid feature because of multicollinearity.
LI-RADS = Liver Imaging Reporting and Data System, HR = hazard ratio, CI = confidence interval, CA = carbohydrate antigen, AJCC = American Joint Committee on Cancer, LR-M = LI-RADS category M
In both models, male sex, perineural invasion, and advanced TNM stage were independently associated with poorer RFS and OS. In Model 1, the presence of at least one CT-based targetoid feature was independently associated with poorer RFS (HR, 2.31; 95% CI, 1.22–4.38; P = 0.010) and OS (HR, 1.85; 95% CI, 1.02–3.36; P = 0.043). The presence of at least one MRI-based non-targetoid feature was also independently associated with poorer RFS (HR, 2.15; 95% CI, 1.26–3.68; P = 0.005) and OS (HR, 2.09; 95% CI, 1.21–3.61; P = 0.008). In Model 2, after adjustment for LI-RADS categorization, MRI-based non-targetoid features remained independently associated with poorer RFS (HR, 1.76; 95% CI, 1.02–3.02; P = 0.041) and OS (HR, 1.79; 95% CI, 1.03–3.12; P = 0.038). In addition, dual-modality LR-M categorization was independently associated with poorer RFS (HR, 8.02; 95% CI, 1.08–59.43; P = 0.042).
Exploratory Analysis of the Association of Individual LR-M Features With Survival
To identify the specific LR-M features associated with survival outcomes, exploratory univariable Cox regression analyses were conducted for individual LR-M features. Among the targetoid features, CT-based rim APHE (HR, 2.00; 95% CI, 1.22–3.28; P = 0.006) and CT-based delayed central enhancement (HR, 1.75; 95% CI, 1.05–2.93; P = 0.033) were significantly associated with poorer RFS (Fig. 4). CT-based rim APHE also showed a significant association with poorer OS (HR, 1.68; 95% CI, 1.04–2.71; P = 0.035). All three MRI-based non-targetoid features, including infiltrative appearance, marked diffusion restriction, and necrosis or severe ischemia, demonstrated significant associations with poorer RFS (HR range, 2.25–2.51; all P < 0.05) and OS (HR range, 2.17–2.51; all P < 0.05). The detailed results are provided in Supplementary Tables 7 and 8.
Fig. 4. CT-based targetoid and MRI-based non-targetoid features in a 38-year-old male with chronic hepatitis B and surgically confirmed intrahepatic cholangiocarcinoma. A: Axial arterial-phase CT image showing a 6-cm mass involving segments V and VI with rim arterial-phase hyperenhancement (arrows). B: Delayed-phase CT image demonstrating delayed central enhancement (arrow). C, D: Axial T1-weighted arterial-phase (C) and hepatobiliary-phase (D) images from gadoxetic acid–enhanced MRI performed on the same day showing a hypoenhancing mass with an infiltrative tumor margin (arrows). E, F: Diffusion-weighted image (E) and apparent diffusion coefficient map (F) demonstrating marked diffusion restriction (arrows) relative to the spleen (arrowheads). Intrahepatic recurrence developed 89 days after margin-negative hepatic resection (not shown), and the patient died of disease progression 473 days after surgery.
Pathologic Characteristics in Relation to CT-Based Targetoid Features and MRI-Based Non-Targetoid Features
iCCAs with at least one CT-based targetoid feature were significantly larger (mean size, 4.8 vs. 3.8 cm; P = 0.005), more frequently demonstrated a purely mass-forming growth pattern (98.9% vs. 87.5%; P = 0.017), and were more often poorly differentiated (23.9% vs. 6.7%; P = 0.040) than those without such features. Tumors with at least one MRI-based non-targetoid feature were significantly larger (mean size, 5.4 vs. 3.9 cm; P = 0.001) and more frequently presented with advanced TNM stage disease (stage III–IV, 39.6% vs. 14.5%; P = 0.005) than those without such features. Detailed comparisons of the pathological characteristics are provided in Supplementary Table 9.
DISCUSSION
In this intraindividual comparison study of patients with resected iCCA at risk of HCC, MRI demonstrated significantly higher frequencies of both targetoid and non-targetoid LR-M features than CT, resulting in a significantly higher rate of LR-M categorization. In the multivariable analysis, the presence of at least one CT-based targetoid feature and at least one MRI-based non-targetoid feature was independently associated with poorer RFS and OS. After accounting for the potential confounding effects of LI-RADS categorization, MRI-based non-targetoid features remained significantly associated with survival outcomes. These findings suggest that the prognostic implications of LR-M features may vary in relation to the imaging modality and feature type and support the potential value of feature-level assessment in iCCA.
Targetoid imaging features are known to reflect the histopathological architecture of iCCA lesions, which consist of a peripheral hypercellular rim and central fibrous stroma [19,20]. Previous studies have consistently shown that iCCAs categorized as LR-M are associated with worse survival outcomes than those categorized as LR-4/5 [8,9,10]. However, these studies assessed the prognostic value of the overall LI-RADS category rather than the individual targetoid features and did not directly compare the CT- and MRI-based features. In the present study, the presence of CT-based targetoid LR-M features was associated with lower survival rates, whereas MRI-based targetoid features showed similar directional trends that did not reach statistical significance. Among individual targetoid features, rim APHE on CT demonstrated the strongest association with RFS (HR, 2.00) and OS (HR, 1.68). Nevertheless, the prognostic significance of CT-based targetoid features should be interpreted with caution because these features are closely linked to LR-M assignment. Indeed, substantial collinearity was observed between the CT-based targetoid features and LI-RADS categorization. Therefore, the observed prognostic association may partly reflect the broader prognostic implications of the LR-M phenotype rather than the isolated effect of individual targetoid features.
MRI-based non-targetoid LR-M features were most consistently associated with adverse survival outcomes, and these associations remained significant even after adjustment for LI-RADS categorization. Each of the three individual MRI-based non-targetoid features independently demonstrated an association with poorer RFS (HR, 2.25–2.51) and poorer OS (HR, 2.17–2.51). Tumors with MRI-based non-targetoid features were significantly larger (mean size, 5.4 vs. 3.9 cm; P = 0.001) and more frequently presented with advanced TNM stage disease (stage III–IV, 39.6% vs. 14.5%; P = 0.005), consistent with a more aggressive tumor phenotype. Among these features, an infiltrative appearance is thought to reflect true tumor cell infiltration of the surrounding liver parenchyma or the confluence of multiple small tumor nodules [6]. This pattern has been reported to be correlated with the large-duct subtype of iCCA, which is associated with poorer postsurgical survival rates than its small-duct subtype counterpart [21]. In our current iCCA cohort, an infiltrative appearance was found to be prognostically significant when assessed using MRI, but not CT. This discrepancy may be attributable to the superior soft-tissue resolution of MRI, particularly on the HBP, which allows more accurate delineation of tumor margins than is possible with CT. Necrosis or severe ischemia, defined as absent or minimal enhancement according to the LI-RADS lexicon, demonstrated prognostic significance on both CT and MRI. This feature is thought to reflect extensive intratumoral ischemia resulting from rapid tumor growth outpacing angiogenesis, a hallmark of aggressive tumor biology. In addition, marked diffusion restriction, which can only be assessed using MRI, has been linked to adverse outcomes. Low apparent diffusion coefficient values have been shown to independently predict worse survival and correlate with stromal desmoplasia and tumor hypoxia, which are features indicative of an aggressive and treatment-resistant tumor microenvironment [22]. Collectively, these findings suggest that MRI-based non-targetoid features may capture the biological characteristics associated with aggressive tumor behavior and provide prognostic information complementary to conventional LI-RADS categorization.
The observed modality-dependent differences may also reflect contrast agent– and phase-related factors. Multiphasic CT uses a purely extracellular contrast agent and provides a true vascular phase, allowing for a more reproducible assessment of tumor vascularity and fibrotic interstitial enhancement. In contrast, most patients in our study cohort underwent MRI using hepatobiliary contrast agents, such as gadoxetic acid (86.1% [105/122]), which are administered in smaller injection volumes and at lower gadolinium doses. These factors may result in weaker arterial enhancement, and the absence of a purely extracellular delayed phase owing to early hepatocyte uptake may reduce the conspicuity and consistency of the enhancement patterns [23]. Therefore, the observed modality-dependent prognostic associations may reflect not only differences in the underlying tumor biology but also differences in feature conspicuity, contrast agent pharmacokinetics, and sensitivity for detecting specific histopathologic components.
Our study had some limitations. First, its retrospective, single-center design may have introduced selection bias. Second, the included patient cohort consisted exclusively of individuals at risk of HCC due to cirrhosis or chronic hepatitis B infection, which may have limited the generalizability of our findings to other iCCA populations. However, this restriction was necessary to ensure strict adherence to the LI-RADS framework, which is specifically intended for patients at risk for HCC. Third, CT and MRI scans were not performed on the same day, and interval tumor progression may have influenced the inter-modality comparisons. However, because most examinations were performed within one month, this effect was likely minimal. Fourth, the MRI examinations were performed using different contrast agents, including hepatobiliary and extracellular agents, which may have affected the conspicuity and prognostic performance of certain MRI features. However, this heterogeneity reflects real-world clinical practice and may enhance the generalizability of our findings. Finally, the relatively small and imbalanced subgroup sizes for certain LR-M feature analyses may have limited statistical power and contributed to imprecise survival estimates. Accordingly, non-significant findings, particularly for MRI-based targetoid and CT-based non-targetoid features, should be interpreted cautiously and warrant validation in larger multicenter cohorts.
In conclusion, the prognostic implications of LR-M imaging features in iCCA may vary in relation to the imaging modality and feature type. CT-based targetoid features were independently associated with poorer survival outcomes, although their prognostic significance was closely linked to the LR-M categorization. MRI-based non-targetoid features demonstrated an independent association with poorer RFS and OS, even after adjustment for LI-RADS categorization. Furthermore, the dual-modality LR-M designation was independently associated with poorer RFS, suggesting additive prognostic value when both CT and MRI indicate LR-M. These findings support the potential value of feature-level and modality-integrated assessments of LR-M observations for imaging-based risk stratification of iCCA.
Footnotes
Conflicts of Interest: Seung Soo Lee, Section Editor of the Korean Journal of Radiology, was not involved in the editorial evaluation or decision to publish this article. The remaining authors have declared no conflicts of interest.
- Conceptualization: Dong Hwan Kim, Sang Hyun Choi.
- Data curation: Yong Jun Jung, Eun Sun Choi, Dong Hwan Kim, Sang Hyun Choi, Seung Soo Lee.
- Formal analysis: Yong Jun Jung, Eun Sun Choi, Dong Hwan Kim, Sang Hyun Choi.
- Investigation: Yong Jun Jung, Dong Hwan Kim, Sang Hyun Choi.
- Methodology: Yong Jun Jung, Dong Hwan Kim, Sang Hyun Choi.
- Project administration: Dong Hwan Kim.
- Resources: Dong Hwan Kim, Sang Hyun Choi.
- Software: Yong Jun Jung, Dong Hwan Kim.
- Supervision: Dong Hwan Kim, Sang Hyun Choi.
- Validation: Yong Jun Jung, Dong Hwan Kim, Sang Hyun Choi.
- Visualization: Yong Jun Jung, Dong Hwan Kim, Sang Hyun Choi.
- Writing—original draft: Yong Jun Jung, Eun Sun Choi, Dong Hwan Kim.
- Writing—review & editing: Sang Hyun Choi, Jae Ho Byun, Hyung Jin Won, Yong Moon Shin.
Funding Statement: None
Availability of Data and Material
The data are not publicly available due to privacy considerations but may be obtained from the corresponding author upon reasonable request.
Supplement
The Supplement is available with this article at https://doi.org/10.3348/kjr.2026.0652.
References
- 1.Bertuccio P, Malvezzi M, Carioli G, Hashim D, Boffetta P, El-Serag HB, et al. Global trends in mortality from intrahepatic and extrahepatic cholangiocarcinoma. J Hepatol. 2019;71:104–114. doi: 10.1016/j.jhep.2019.03.013. [DOI] [PubMed] [Google Scholar]
- 2.Clements O, Eliahoo J, Kim JU, Taylor-Robinson SD, Khan SA. Risk factors for intrahepatic and extrahepatic cholangiocarcinoma: a systematic review and meta-analysis. J Hepatol. 2020;72:95–103. doi: 10.1016/j.jhep.2019.09.007. [DOI] [PubMed] [Google Scholar]
- 3.Tsilimigras DI, Sahara K, Wu L, Moris D, Bagante F, Guglielmi A, et al. Very early recurrence after liver resection for intrahepatic cholangiocarcinoma: considering alternative treatment approaches. JAMA Surg. 2020;155:823–831. doi: 10.1001/jamasurg.2020.1973. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Choi WJ, Williams PJ, Claasen MPAW, Ivanics T, Englesakis M, Gallinger S, et al. Systematic review and meta-analysis of prognostic factors for early recurrence in intrahepatic cholangiocarcinoma after curative-intent resection. Ann Surg Oncol. 2022;29:4337–4353. doi: 10.1245/s10434-022-11463-x. [DOI] [PubMed] [Google Scholar]
- 5.de Jong MC, Nathan H, Sotiropoulos GC, Paul A, Alexandrescu S, Marques H, et al. Intrahepatic cholangiocarcinoma: an international multi-institutional analysis of prognostic factors and lymph node assessment. J Clin Oncol. 2011;29:3140–3145. doi: 10.1200/JCO.2011.35.6519. [DOI] [PubMed] [Google Scholar]
- 6.American College of Radiology. LI-RADS® CT/MRI v2018 core. [accessed on December 30, 2025]. Available at: https://edge.sitecorecloud.io/americancoldf5f-acrorgf92a-productioncb02-3650/media/ACR/Files/RADS/LI-RADS/LI-RADS-CT-MRI-2018-Core.pdf .
- 7.Kim DH, Choi SH, Park SH, Kim KW, Byun JH, Kim SY, et al. Liver imaging reporting and data system category M: a systematic review and meta-analysis. Liver Int. 2020;40:1477–1487. doi: 10.1111/liv.14420. [DOI] [PubMed] [Google Scholar]
- 8.Park S, Koo B, Jeong B, Choi SH, Lee JM. LI-RADS category can be a post-surgical prognostic factor for intrahepatic cholangiocarcinoma in patients with liver cirrhosis or chronic hepatitis B. Liver Cancer. 2024;13:629–642. doi: 10.1159/000539794. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Hwang JA, Lee S, Lee JE, Yoon J, Choi SY, Shin J. LI-RADS category on MRI is associated with recurrence of intrahepatic cholangiocarcinoma after surgery: a multicenter study. J Magn Reson Imaging. 2023;57:930–938. doi: 10.1002/jmri.28354. [DOI] [PubMed] [Google Scholar]
- 10.Sheng R, Zheng B, Zhang Y, Yang C, Wu D, Zhou J, et al. Assessment of intrahepatic cholangiocarcinoma with LI-RADS in the high-risk population: MRI diagnosis and postoperative survival. Cancer Imaging. 2025;25:40. doi: 10.1186/s40644-025-00860-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Corwin MT, Fananapazir G, Jin M, Lamba R, Bashir MR. Differences in liver imaging and reporting data system categorization between MRI and CT. AJR Am J Roentgenol. 2016;206:307–312. doi: 10.2214/AJR.15.14788. [DOI] [PubMed] [Google Scholar]
- 12.Chernyak V, Flusberg M, Law A, Kobi M, Paroder V, Rozenblit AM. Liver imaging reporting and data system: discordance between computed tomography and gadoxetate-enhanced magnetic resonance imaging for detection of hepatocellular carcinoma major features. J Comput Assist Tomogr. 2018;42:155–161. doi: 10.1097/RCT.0000000000000642. [DOI] [PubMed] [Google Scholar]
- 13.Zhang YD, Zhu FP, Xu X, Wang Q, Wu CJ, Liu XS, et al. Liver imaging reporting and data system: substantial discordance between CT and MR for Imaging classification of hepatic nodules. Acad Radiol. 2016;23:344–352. doi: 10.1016/j.acra.2015.11.002. [DOI] [PubMed] [Google Scholar]
- 14.An C, Lee CH, Byun JH, Lee MH, Jeong WK, Choi SH, et al. Intraindividual comparison between gadoxetate-enhanced magnetic resonance imaging and dynamic computed tomography for characterizing focal hepatic lesions: a multicenter, multireader study. Korean J Radiol. 2019;20:1616–1626. doi: 10.3348/kjr.2019.0363. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.European Association for the Study of the Liver. EASL-ILCA clinical practice guidelines on the management of intrahepatic cholangiocarcinoma. J Hepatol. 2023;79:181–208. doi: 10.1016/j.jhep.2023.03.010. [DOI] [PubMed] [Google Scholar]
- 16.Moris D, Palta M, Kim C, Allen PJ, Morse MA, Lidsky ME. Advances in the treatment of intrahepatic cholangiocarcinoma: an overview of the current and future therapeutic landscape for clinicians. CA Cancer J Clin. 2023;73:198–222. doi: 10.3322/caac.21759. [DOI] [PubMed] [Google Scholar]
- 17.Nagtegaal ID, Odze RD, Klimstra D, Paradis V, Rugge M, Schirmacher P, et al. The 2019 WHO classification of tumours of the digestive system. Histopathology. 2020;76:182–188. doi: 10.1111/his.13975. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Lee AJ, Chun YS. Intrahepatic cholangiocarcinoma: the AJCC/UICC 8th edition updates. Chin Clin Oncol. 2018;7:52. doi: 10.21037/cco.2018.07.03. [DOI] [PubMed] [Google Scholar]
- 19.Vijgen S, Terris B, Rubbia-Brandt L. Pathology of intrahepatic cholangiocarcinoma. Hepatobiliary Surg Nutr. 2017;6:22–34. doi: 10.21037/hbsn.2016.11.04. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Seo N, Kim DY, Choi JY. Cross-sectional imaging of intrahepatic cholangiocarcinoma: development, growth, spread, and prognosis. AJR Am J Roentgenol. 2017;209:W64–W75. doi: 10.2214/AJR.16.16923. [DOI] [PubMed] [Google Scholar]
- 21.Park S, Lee Y, Kim H, Yu MH, Lee ES, Yoon JH, et al. Subtype classification of intrahepatic cholangiocarcinoma using liver MR imaging features and its prognostic value. Liver Cancer. 2022;11:233–246. doi: 10.1159/000521747. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Yamada S, Morine Y, Imura S, Ikemoto T, Arakawa Y, Saito Y, et al. Prognostic prediction of apparent diffusion coefficient obtained by diffusion-weighted MRI in mass-forming intrahepatic cholangiocarcinoma. J Hepatobiliary Pancreat Sci. 2020;27:388–395. doi: 10.1002/jhbp.732. [DOI] [PubMed] [Google Scholar]
- 23.Huh J, Kim SY, Yeh BM, Lee SS, Kim KW, Wu EH, et al. Troubleshooting arterial-phase MR images of gadoxetate disodium-enhanced liver. Korean J Radiol. 2015;16:1207–1215. doi: 10.3348/kjr.2015.16.6.1207. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data are not publicly available due to privacy considerations but may be obtained from the corresponding author upon reasonable request.