Abstract
Purpose
The present study aimed to develop and validate a preoperative model based on Gd-DTPA-enhanced magnetic resonance imaging (MRI) and clinical factors for predicting perineural invasion (PNI) in patients with intrahepatic cholangiocarcinoma (ICC), enabling clinicians to perform more accurate patient evaluation and and make individualized therapeutic decisions.
Methods
Between July 2019 and February 2024, a total of 173 patients with pathologically confirmed ICC who underwent preoperative Gd‑DTPA‑enhanced MRI were retrospectively enrolled. These patients were randomly assigned to training and test cohorts at a 7:3 ratio. Multivariate logistic regression was used to identify independent predictors of PNI status and a predictive model was developed and presented as a nomogram. The model performance was assessed in terms of discrimination, calibration, and clinical utility.
Results
Peritumoral arterial hyper-enhancement (OR 14.99, 95% CI:1.95,115.22, P = 0.009), satellite nodules (OR 10.07, 95% CI: 1.34,75.48, P = 0.025), target sign on DWI (OR 0.03, 95% CI: 0.00,0.38, P = 0.007), and tumor location (OR 0.03, 95% CI: 0.00,0.28, P = 0.003) were independent risk factors to PNI status and constituted the model. The model exhibited strong discriminative ability, with an area under the curve (AUC) of 0.939 (95% CI: 0.891–0.988), sensitivity of 0.907, specificity of 0.870, and accuracy of 0.893 in the training set; corresponding values in the testing set were 0.866 (95% CI: 0.767–0.966), 0.750, 0.800, and 0.769. The decision curve analysis (DCA) curve showed that the model achieved great clinical benefits.
Conclusion
In conclusion, we developed and validated a noninvasive preoperative model that integrates Gd-DTPA-enhanced MRI features to accurately predict PNI in ICC patients. This model shows significant potential to assist clinicians in preoperative risk assessment, thus facilitating improved prognostic stratification and the formulation of personalized treatment strategies.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12880-026-02188-4.
Keywords: Intrahepatic cholangiocarcinoma, Perineural invasion, Magnetic resonance imaging, Nomogram
Introduction
Intrahepatic cholangiocarcinoma (ICC) is a highly aggressive malignancy arising from the intrahepatic bile duct epithelium and is the second most common primary liver tumor, accounting for approximately 10%-15% of primary hepatic malignancies [1, 2]. Surgical resection is the cornerstone of potentially curative treatment [3, 4]. Nevertheless, this is frequently thwarted by a high postoperative recurrence rate, which significantly compromises long-term survival [5].
Perineural invasion (PNI), defined histopathologically as tumor cells within the perineurium, surrounding nerves, or invading neural structures (6), is an established independent prognostic factor for poor outcomes, including higher recurrence, metastasis, and worse survival [6, 7]. However, current PNI assessment depends solely on postoperative histopathology, which precludes preoperative risk stratification and surgical planning [8, 9]. As preoperative knowledge of PNI status could inform critical decisions—such as resection extent or neoadjuvant therapy—its accurate prediction is crucial for formulating individualized treatment strategies and improving prognosis in ICC patients [10, 11].
Magnetic resonance imaging (MRI), with its superior soft-tissue contrast and multiplanar capabilities, has emerged as a pivotal tool in the preoperative evaluation of ICC [12]. And it is very critical in clinical and research applications.Given the intricate relationship between tumor biology and imaging phenotype, it is plausible that specific MRI features—such as tumor margin irregularity, peritumoral enhancement, or neural structure involvement— may serve as surrogates for PNI [13].
While recent studies have explored image-based models for predicting PNI in ICC, including CT-based radiomics models and multiparametric MRI radiomics approaches [14, 15], most of these rely on computationally extracted radiomic features that require specialized software and may lack clinical interpretability. In contrast, our model is based on conventional, visually-assessed MRI features, which are more readily interpretable and easier to implement in routine clinical practice. Additionally, although previous studies have investigated the association between imaging features and PNI [16], few have integrated multiple Gd‑DTPA‑enhanced MRI features into a predictive nomogram for preoperative PNI assessment in ICC.
Therefore, the aim of this study was to develop and validate a preoperative model based on Gd-DTPA-enhanced MRI and clinical factors for predicting PNI status in ICC patients, enabling clinicians to perform more accurate patient evaluation and and make individualized therapeutic decisions.
Materials and methods
Patients
Between July 2019 and February 2024, 173 patients with pathologically confirmed ICC who underwent preoperative Gd-DTPA-enhanced MRI were initially enrolled Fig. 1. The inclusion criteria were: (a) complete histopathological confirmation of ICC; (b) dynamic contrast-enhanced liver MRI performed within one month before surgery, including arterial, portal, and delayed phase images. Exclusion criteria were: (a) preoperative adjuvant tumor therapy such as radiotherapy; (b) unidentifiable primary tumor or poor image quality on MRI; or (c) incomplete clinical or imaging data. Eligible patients were randomly allocated to a training cohort (n = 121) and a test cohort (n = 52) at a 7:3 ratio using a computer-generated random number sequence. The training cohort was used for model development, and the test cohort served for internal validation.
Fig. 1.
The workflow of patient selection for this study
Laboratory examinations and histopathology
Preoperative laboratory indexes (Table 1) comprised protein induced by vitamin K absence/antagonist-II (PIVKA-II), serum alpha-fetoprotein (AFP), carbohydrate antigen 19 − 9 (CA19-9), carcinoembryonic antigen (CEA), cholinesterase (CHE), total bilirubin, direct bilirubin, r-glutamyltransferase,α-L-fucosidase, albumin (ALB), total cholesterol, prothrombin time, etc.
Table 1.
Baseline characteristics of patients with ICC in training and testing cohort
| Characteristic | Training cohort (n = 121) | Testing cohort (n = 52) | ||||
|---|---|---|---|---|---|---|
| PNI- (n = 75) | PNI+ (n = 46) | P value | PNI- (n = 32) | PNI+ (n = 20) | P value | |
| Clinical features | ||||||
| Age, M (Q₁, Q₃),year | 58.00 (51.50, 64.50) | 59.00 (52.25, 68.00) | 0.331 | 59.00 (50.75, 65.25) | 58.00 (54.00, 65.25) | 0.541 |
| Sex, n(%) | 0.773 | 0.779 | ||||
| Male | 46 (61.33) | 27 (58.70) | 22 (68.75) | 13 (65.00) | ||
| Female | 29 (38.67) | 19 (41.30) | 10 (31.25) | 7 (35.00) | ||
| Liver disease, n(%) | 0.026 | 0.508 | ||||
| HBV/HCV | 33 (44.00) | 11 (23.91) | 13 (40.62) | 10 (50.00) | ||
| None | 42 (56.00) | 35 (76.09) | 19 (59.38) | 10 (50.00) | ||
| Microscopic cirrhosis, n(%) | 0.635 | 0.864 | ||||
| absent | 70 (93.33) | 41 (89.13) | 29 (90.62) | 17 (85.00) | ||
| present | 5 (6.67) | 5 (10.87) | 3 (9.38) | 3 (15.00) | ||
| PIVKA-II, M (Q₁, Q₃),(mAU/mL) | 22.50 (19.00, 26.75) | 21.50 (18.00, 26.00) | 0.361 | 21.00 (19.00, 27.00) | 24.00 (20.50, 31.75) | 0.298 |
| AFP, M (Q₁, Q₃),(µg/L) | 3.20 (2.24, 6.55) | 3.65 (2.59, 5.10) | 0.586 | 3.80 (1.85, 4.75) | 3.15 (2.60, 4.28) | 0.764 |
| CEA, M (Q₁, Q₃),(µg/ml) | 2.57 (1.70, 3.62) | 4.40 (2.31, 6.33) | < 0.001 | 2.01 (1.65, 2.80) | 3.15 (2.42, 4.68) | 0.015 |
| CA199, M (Q₁, Q₃),(U/mL) | 24.60 (11.60, 96.19) | 326.50 (38.78, 1000.00) | < 0.001 | 21.60 (12.90, 83.80) | 209.10 (80.21, 1000.00) | 0.004 |
| CA125, M (Q₁, Q₃),(U/mL) | 13.10 (9.80, 20.44) | 20.07 (9.55, 51.10) | 0.075 | 17.80 (12.65, 28.57) | 23.60 (9.80, 43.80) | 0.660 |
| TBIL, M (Q₁, Q₃),(µmol/L) | 12.80 (9.40, 14.95) | 10.55 (8.12, 15.50) | 0.242 | 12.65 (8.70, 14.70) | 16.65 (10.82, 23.47) | 0.098 |
| DBIL, M (Q₁, Q₃),(µmol/L) | 4.60 (3.50, 5.80) | 4.15 (2.90, 5.77) | 0.346 | 4.15 (3.68, 5.62) | 5.65 (3.85, 9.43) | 0.094 |
| IBIL, M (Q₁, Q₃),(µmol/L) | 7.60 (5.85, 9.45) | 6.65 (5.20, 8.73) | 0.120 | 7.40 (4.97, 10.22) | 8.65 (6.70, 13.07) | 0.188 |
| CG, M (Q₁, Q₃),(µg/L) | 1.10 (0.60, 1.50) | 1.45 (0.83, 3.08) | 0.037 | 1.40 (0.90, 2.50) | 2.15 (0.67, 10.43) | 0.440 |
| ALB, M (Q₁, Q₃),(g/L) | 42.50 (40.60, 44.55) | 42.05 (39.95, 43.82) | 0.196 | 40.55 (39.05, 43.68) | 41.85 (38.92, 42.85) | 0.985 |
| GLOB, M (Q₁, Q₃),(g/L) | 26.50 (23.75, 29.10) | 27.90 (24.45, 31.30) | 0.091 | 25.25 (23.48, 27.18) | 25.45 (22.62, 29.82) | 0.771 |
| CHE, M (Q₁, Q₃),(U/L) | 7546.50 (6506.50, 8248.75) | 7272.00 (6343.00, 8348.00) | 0.902 | 7044.00 (6256.50, 8715.00) | 6964.00 (6049.00, 8000.00) | 0.519 |
| ALT, M (Q₁, Q₃),(U/L) | 21.00 (16.00, 34.00) | 23.50 (18.25, 39.75) | 0.204 | 21.00 (12.75, 30.00) | 26.00 (18.00, 81.50) | 0.051 |
| AST, M (Q₁, Q₃),(U/L) | 23.00 (18.00, 30.00) | 26.00 (20.00, 36.75) | 0.084 | 19.00 (15.00, 27.00) | 32.00 (22.50, 46.00) | 0.005 |
| GGT, M (Q₁, Q₃),(U/L) | 43.00 (21.00, 92.00) | 90.50 (60.75, 178.75) | < 0.001 | 43.00 (27.00, 87.00) | 186.00 (64.50, 332.50) | < 0.001 |
| AFU, M (Q₁, Q₃),(U/L) | 24.00 (19.00, 28.20) | 26.00 (20.00, 31.00) | 0.234 | 25.80 (20.05, 30.70) | 30.45 (22.88, 38.22) | 0.184 |
| CHOL, M (Q₁, Q₃),(mmol/L) | 4.05 (3.64, 4.41) | 4.12 (3.76, 5.08) | 0.147 | 4.03 (3.43, 4.45) | 4.41 (3.79, 5.23) | 0.387 |
| TG, M (Q₁, Q₃),(mmol/L) | 1.12 (0.87, 1.45) | 1.23 (0.92, 1.54) | 0.525 | 1.11 (0.91, 2.06) | 1.38 (1.06, 1.94) | 0.556 |
| HDL, M (Q₁, Q₃),(mmol/L) | 1.10 (0.93, 1.38) | 1.17 (0.95, 1.35) | 0.713 | 1.12 (0.95, 1.28) | 0.94 (0.81, 1.23) | 0.321 |
| LDL, M (Q₁, Q₃),(mmol/L) | 2.56 (2.15, 3.03) | 2.70 (2.09, 3.34) | 0.322 | 2.29 (1.78, 2.88) | 2.43 (1.99, 2.60) | 0.819 |
| CRP, M (Q₁, Q₃),(mg/L) | 1.37 (0.50, 4.78) | 1.73 (0.71, 9.82) | 0.108 | 1.55 (0.50, 5.55) | 2.47 (0.86, 10.64) | 0.266 |
| PLT, M (Q₁, Q₃),(10^9/L) | 193.50 (148.50, 228.25) | 200.50 (152.50, 249.50) | 0.330 | 191.50 (161.75, 239.25) | 210.00 (159.75, 229.25) | 0.933 |
| PT, M (Q₁, Q₃),(S) | 11.50 (11.00, 11.90) | 11.25 (11.00, 11.60) | 0.348 | 11.55 (11.00, 11.83) | 11.60 (11.10, 12.20) | 0.355 |
| APTT, M (Q₁, Q₃),(S) | 26.65 (24.50, 28.08) | 26.75 (25.52, 29.00) | 0.519 | 27.15 (25.25, 28.30) | 26.25 (25.45, 27.45) | 0.457 |
| TT, M (Q₁, Q₃),(S) | 18.10 (17.02, 19.70) | 18.30 (17.17, 19.12) | 0.993 | 18.70 (17.85, 19.65) | 19.00 (17.65, 19.52) | 1.000 |
| MRI features | ||||||
| Largest diameter, M (Q₁, Q₃) | 4.80 (3.30, 6.25) | 5.85 (3.28, 7.25) | 0.076 | 5.45 (4.10, 7.35) | 4.60 (3.18, 6.78) | 0.547 |
| Tumor location, n(%) | < 0.001 | < 0.001 | ||||
| perihilar | 9 (12.00) | 37 (80.43) | 5 (15.62) | 14 (70.00) | ||
| subcapsular | 66 (88.00) | 9 (19.57) | 27 (84.38) | 6 (30.00) | ||
| Morphologic, n(%) | 0.084 | 0.641 | ||||
| ill-defined | 2 (2.67) | 6 (13.04) | 2 (6.25) | 3 (15.00) | ||
| lobulated | 38 (50.67) | 23 (50.00) | 12 (37.50) | 6 (30.00) | ||
| round | 35 (46.67) | 17 (36.96) | 18 (56.25) | 11 (55.00) | ||
| Margin, n(%) | 0.005 | 0.004 | ||||
| no-smooth | 26 (34.67) | 28 (60.87) | 11 (34.38) | 15 (75.00) | ||
| smooth | 49 (65.33) | 18 (39.13) | 21 (65.62) | 5 (25.00) | ||
| SI on T2WI, n(%) | 0.131 | 0.249 | ||||
| heterogeneous | 21 (28.00) | 19 (41.30) | 13 (40.62) | 5 (25.00) | ||
| homogeneous | 54 (72.00) | 27 (58.70) | 19 (59.38) | 15 (75.00) | ||
| SI on T1WI, n(%) | 0.061 | 1.000 | ||||
| heterogeneous | 6 (8.00) | 9 (19.57) | 4 (12.50) | 2 (10.00) | ||
| homogeneous | 69 (92.00) | 37 (80.43) | 28 (87.50) | 18 (90.00) | ||
| Surface retraction, n(%) | 0.439 | 0.476 | ||||
| absence | 47 (62.67) | 32 (69.57) | 21 (65.62) | 15 (75.00) | ||
| presence | 28 (37.33) | 14 (30.43) | 11 (34.38) | 5 (25.00) | ||
| Bile duct dilatation, n(%) | < 0.001 | < 0.001 | ||||
| absence | 61 (81.33) | 17 (36.96) | 26 (81.25) | 4 (20.00) | ||
| presence | 14 (18.67) | 29 (63.04) | 6 (18.75) | 16 (80.00) | ||
| Intrallesional fat, n(%) | 0.706 | 1.000 | ||||
| absence | 71 (94.67) | 45 (97.83) | 31 (96.88) | 20 (100.00) | ||
| presence | 4 (5.33) | 1 (2.17) | 1 (3.12) | 0 (0.00) | ||
| Capsule, n(%) | 0.319 | 1.000 | ||||
| absence | 64 (85.33) | 36 (78.26) | 26 (81.25) | 17 (85.00) | ||
| presence | 11 (14.67) | 10 (21.74) | 6 (18.75) | 3 (15.00) | ||
| Target sign on DWI, n(%) | 0.002 | 0.660 | ||||
| absence | 23 (30.67) | 27 (58.70) | 14 (43.75) | 10 (50.00) | ||
| presence | 52 (69.33) | 19 (41.30) | 18 (56.25) | 10 (50.00) | ||
| PV embolus, n(%) | 0.003 | 0.977 | ||||
| absence | 69 (92.00) | 33 (71.74) | 27 (84.38) | 16 (80.00) | ||
| presence | 6 (8.00) | 13 (28.26) | 5 (15.62) | 4 (20.00) | ||
| Septum, n(%) | 0.006 | 0.319 | ||||
| absence | 67 (89.33) | 32 (69.57) | 26 (81.25) | 19 (95.00) | ||
| presence | 8 (10.67) | 14 (30.43) | 6 (18.75) | 1 (5.00) | ||
| Lymph node metastasis, n(%) | 0.006 | 0.254 | ||||
| absence | 60 (80.00) | 26 (56.52) | 24 (75.00) | 12 (60.00) | ||
| presence | 15 (20.00) | 20 (43.48) | 8 (25.00) | 8 (40.00) | ||
| Intrahepatic bile duct calculus, n(%) | 0.004 | 0.672 | ||||
| absence | 74 (98.67) | 38 (82.61) | 31 (96.88) | 18 (90.00) | ||
| presence | 1 (1.33) | 8 (17.39) | 1 (3.12) | 2 (10.00) | ||
| Satellite nodules, n(%) | < 0.001 | 0.048 | ||||
| absence | 63 (84.00) | 25 (54.35) | 27 (84.38) | 12 (60.00) | ||
| presence | 12 (16.00) | 21 (45.65) | 5 (15.62) | 8 (40.00) | ||
| Arterial phase enhancement, n(%) | < 0.001 | < 0.001 | ||||
| Diffuse hyper-enhancement | 10 (13.33) | 0 (0.00) | 4 (12.50) | 2 (10.00) | ||
| Diffuse hypo-enhancement | 0 (0.00) | 15 (33.33) | 0 (0.00) | 8 (40.00) | ||
| Peripheral rim | 65 (86.67) | 30 (66.67) | 28 (87.50) | 10 (50.00) | ||
| Dynamic enhancement pattern, n(%) | < 0.001 | 1.000 | ||||
| Persistent hyper-enhancement | 10 (13.33) | 0 (0.00) | 4 (12.50) | 2 (10.00) | ||
| Persistent hypo-enhancement | 1 (1.33) | 7 (15.56) | 1 (3.12) | 1 (5.00) | ||
| Progressive interstitial | 64 (85.33) | 38 (84.44) | 27 (84.38) | 17 (85.00) | ||
| Peritumoral arterial hyper, n(%) | < 0.001 | 0.002 | ||||
| absence | 59 (78.67) | 13 (28.26) | 25 (78.12) | 7 (35.00) | ||
| presence | 16 (21.33) | 33 (71.74) | 7 (21.88) | 13 (65.00) | ||
| Z: Mann-Whitney test, χ²: Chi-square test, Fisher exact | ||||||
M: Median, Q₁: 1st Quartile, Q₃: 3st Quartile
Abbreviations: PNI, perineural invasion; HBV, hepatitis B virus; AFP, alpha-fetoprotein; PIVKA-II, protein induced by vitamin K absence or antagonist-II; CA199, carbohydrate antigen 19 − 9; CEA, carcinoembryonic antigen; ALT, alanine aminotransferase; AST, aspartate aminotransferase; TBIL, total bilirubin; DBIL, direct bilirubin; CHE, cholinesterase; ALB, albumin; GLOB, globulin; GGT, r-glutamyl transferase; AFU, a-fucosidase; PT, prothrombin time; CHOL, total cholesterol; APHE, arterial phase hyperenhancement; HBP, hepatobiliary phase
The histopathological evaluation was performed by two experienced pathologists, who reached a consensus on all cases. Perineural invasion (PNI) was diagnosed based on established criteria, defined as the microscopic presence of tumor cells infiltrating any layer of the nerve sheath (epineurium, perineurium, or endoneurium) or encircling ≥ 33% of the nerve circumference [17]. According to this definition, patients were categorized into PNI-positive (+) or PNI-negative (−) groups.
MRI examination
MR imaging was performed on a 1.5T scanner (GE Optima MR360) using an eight-channel abdominal coil. After a four-hour fast, patients underwent baseline sequences including in- and opposed-phase T1-weighted imaging (T1WI) and fat-suppressed T2-weighted imaging (Fs-T2WI). Diffusion-weighted imaging (DWI) was acquired using a respiratory-triggered single-shot echo-planar sequence with b-values of 0 and 600 s/mm².
Gadolinium meglumine acid (Gd-DTPA) was injected at 0.1 mmol/kg via the median cubital vein (2.0 mL/s) and flushed with 20 mL saline. Dynamic imaging included arterial, portal venous, and delayed phases at 20–30 s, 50–60 s, and 90–120 s, respectively, with full parameters in Supplementary Table 1.
MR imaging analysis
Two radiologists, each with over ten years of experience in abdominal imaging and blinded to histopathological results, independently performed the MR image analysis. Any discrepancies in their evaluations were resolved through discussion to reach a consensus.
Magnetic resonance imaging features were assessed according to the following categories:
General morphological features
(a) Tumor size: measured on the largest cross-sectional plane in the portal venous phase, including the capsule if present, following LI-RADS v2018 criteria [18];(b) Shape: classified as round, lobulated, or ill-defined; (c) Margin: categorized as smooth or non-smooth, the latter referring to nodular or budding contours on axial or coronal images; (d) Signal intensity: evaluated as homogeneous or heterogeneous on T1WI and T2WI; (e) Location: recorded as subcapsular or perihilar.
Contrast enhancement features
(a) Arterial phase enhancement; (b) Dynamic enhancement pattern; (c) Peritumoral arterial hyperenhancement.
Ancillary imaging features
Peripheral bile duct dilatation; (b) Hepatic surface retraction adjacent to the tumor; (c) Capsule appearance, defined as partial or complete rim-like enhancement on portal venous or delayed phases; (d) Intrahepatic bile duct calculus; (e) Septum in the tumor; (f) Target sign on DWI (b = 600 s/mm²); (g) Intralesional fat; (h) Portal vein embolus; (i) Lymph node metastasis; (j) Satellite nodules.
A comprehensive description of all imaging feature definitions and evaluation criteria is provided in the Supplementary Methods.
Model development, validation, and evaluation
Patients were randomly assigned to training and test cohorts at a 7:3 ratio. Independent predictors of PNI status were identified using multivariate logistic regression. These predictors were incorporated into a nomogram to establish the prediction model. The model’s performance was assessed in terms of discrimination, calibration, and clinical utility. Discrimination was measured by the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. Calibration was evaluated using calibration curves to examine the agreement between predicted probabilities and observed outcomes. Clinical utility was assessed via decision curve analysis (DCA), which quantified the net benefit across a range of threshold probabilities.
Statistical analysis
Statistical analyses were performed using IBM SPSS Statistics (version 25) or R (version 4.4.1; http://www.r-project.org). Normally distributed continuous variables were expressed as mean ± standard deviation and compared using the Student’s t-test, while non-normally distributed variables were summarized as median (range) and analyzed with the Mann–Whitney U test. Categorical variables were compared using the χ2 test or Fisher’s exact test, as appropriate. Interobserver agreement between radiologists was assessed with the kappa statistic, and variables with kappa values below 0.75 were excluded from further analysis. Variables showing a significance level of P < 0.05 in univariate logistic regression were entered into multivariate logistic regression to identify independent predictors for model construction. Prior to multivariate analysis, multicollinearity among candidate variables was assessed using the variance inflation factor (VIF).
Results
Baseline characteristics of the study cohorts
Among the 173 patients (mean age, 58.2 ± 9.9 years;108 male and 65 female), 66 (38.2%) were diagnosed with PNI(+). Of these, PNI incidence rates were 38.0% (46/121), and 38.5%(20/52) in training and test cohort, respectively. The clinical and MR imaging features of the different cohorts were presented in Table 1. The two radiologists (Readers 1 and 2) showed a consistent analysis of qualitative MRI features, as the kappa values ≥ 0.75.The equilibrium test for training and test sets were presented in Supplement Table 2.
Table 2.
Univariate analyses for predicting PNI in the training cohort
| Variables | β | OR (95%CI) | P |
|---|---|---|---|
| Liver disease | |||
| HBV/HCV | 1.00 (Reference) | ||
| None | 0.92 | 2.50 (1.10 ~ 5.66) | 0.028 |
| Tumor location | |||
| perihilar | 1.00 (Reference) | ||
| subcapsular | -3.41 | 0.03 (0.01 ~ 0.09) | < 0.001 |
| Morphologic | |||
| ill-defined | 1.00 (Reference) | ||
| lobulated | -1.60 | 0.20 (0.04 ~ 1.08) | 0.062 |
| round | -1.82 | 0.16 (0.03 ~ 0.89) | 0.036 |
| Margin | |||
| no-smooth | 1.00 (Reference) | ||
| smooth | -1.08 | 0.34 (0.16 ~ 0.73) | 0.006 |
| Bile duct dilatation | |||
| absence | 1.00 (Reference) | ||
| presence | 2.01 | 7.43 (3.23 ~ 17.12) | < 0.001 |
| Target sign on DWI | |||
| absence | 1.00 (Reference) | ||
| presence | -1.17 | 0.31 (0.14 ~ 0.67) | 0.003 |
| PV embolus | |||
| absence | 1.00 (Reference) | ||
| presence | 1.51 | 4.53 (1.58 ~ 12.98) | 0.005 |
| Septum | |||
| absence | 1.00 (Reference) | ||
| presence | 1.30 | 3.66 (1.40 ~ 9.62) | 0.008 |
| Lymph node metastasis | |||
| absence | 1.00 (Reference) | ||
| presence | 1.12 | 3.08 (1.37 ~ 6.93) | 0.007 |
| Intrahepatic bile duct calculus | |||
| absence | 1.00 (Reference) | ||
| presence | 2.75 | 15.58 (1.88 ~ 129.18) | 0.011 |
| Satellite nodules | |||
| absence | 1.00 (Reference) | ||
| presence | 1.48 | 4.41 (1.89 ~ 10.29) | < 0.001 |
| Peritumoral arterial hyper | |||
| absence | 1.00 (Reference) | ||
| presence | 2.24 | 9.36 (4.01 ~ 21.83) | < 0.001 |
| CA199 | 0.01 | 1.01 (1.01 ~ 1.01) | < 0.001 |
| GGT | 0.01 | 1.01 (1.01 ~ 1.01) | 0.020 |
OR: Odds Ratio, CI: Confidence Interval.Abbreviations can be found in the notes of Table 1
Univariate and multivariate analysis factors predictive of PNI
Univariate analysis showed tumor location, tumour morphology, margin, bile duct dilatation, target sign on DWI, PV embolus, septum, lymph node metastasis, intrahepatic bile duct calculus, satellite nodules, peritumoral arterial hyper-enhancement, liver disease, CA19-9, and GGT levels were significantly related to the PNI (P < 0.05,Table 2). The CA19-9 (P<0.001) and GGT (P = 0.020) levels in the PNI(+) group were higher than those in the PNI(-) group.
Among the MRI features, tumor located in the perihilar (80.43% VS. 12.00%, P < 0.001), non-smooth margin (60.87% vs. 34.67%, P = 0.005), bile duct dilatation (63.04% VS. 18.67%, P < 0.001), PV embolus (28.26% VS. 8.00%, P = 0.003), septum (30.43% VS. 10.67%, P = 0.006), lymph node metastasis (43.48% VS. 20.00%, P = 0.006), intrahepatic bile duct calculus (17.39% VS. 1.33%, P = 0.004), satellite nodules (45.65% VS. 16.00%, P < 0.001), peritumoral arterial hyper-enhancement (71.74% VS. 21.33%, P < 0.001) had a higher probability in the PNI positive group than in the PNI negative group.
Multicollinearity assessment among the 14 significant univariate variables revealed no VIF > 5, indicating acceptable levels of collinearity for inclusion in the multivariate model (Supplementary Table 3). In multivariate logistic regression analysis, four features were identified as independent predictors: tumor location (odds ratio (OR) 0.03, 95% confidence interval (CI): 0.00, 0.28, P = 0.003), target sign on DWI (OR 0.03, 95%CI: 0.00,0.38, P = 0.007), satellite nodules (OR 10.07, 95%CI: 1.34,75.48, P = 0.025), and peritumoral arterial hyper-enhancement (OR 14.99, 95%CI: 1.95,115.22, P = 0.009) (Table 3). The very low ORs for tumor location (perihilar vs. subcapsular) and target sign on DWI indicate a strong protective association. Therefore, the above four risk factors constituted the model for predicting PNI in patients with ICC.The representative images of PNI positive case is displayed in Fig. 2, and PNI negative case is displayed in Fig. 3.
Table 3.
Multivariate analyses for predicting PNI in the training cohort
| Variables | β | OR (95%CI) | P |
|---|---|---|---|
| Tumor location | |||
| perihilar | 1.00 (Reference) | ||
| subcapsular | -3.66 | 0.03 (0.00 ~ 0.28) | 0.003 |
| Target sign on DWI | |||
| absence | 1.00 (Reference) | ||
| presence | -3.55 | 0.03 (0.00 ~ 0.38) | 0.007 |
| Satellite nodules | |||
| absence | 1.00 (Reference) | ||
| presence | 2.31 | 10.07 (1.34 ~ 75.48) | 0.025 |
| Peritumoral arterial hyper | |||
| absence | 1.00 (Reference) | ||
| presence | 2.71 | 14.99 (1.95 ~ 115.22) | 0.009 |
OR: Odds Ratio, CI: Confidence Interval
Fig. 2.
Representative images of PNI positive case (A-F):Gd- DPTA MRI detected a perihilar lesion (A, E), bile duct dilatation (B), restricted diffusion and no target sign (C), atypical enhancement pattern with progressive interstitial (D-F) and peritumoral enhancement on arterial phase images (D)
Fig. 3.
Representative images of PNI negative case (A-F): Gd- DPTA MRI detected a subcapsular lesion (A), surface retraction (B, E),without bile duct dilatation (B), target sign on DWI (C), atypical enhancement pattern with progressive interstitial (D-F), without peritumoral enhancement on arterial phase images (D)
Model development, validation, and evaluation
A nomogram for predicting PNI in ICC patients was developed and is presented in Fig. 4A. The model exhibited strong discriminative ability, with an area under the curve (AUC) of 0.939 (95% CI: 0.891–0.988), sensitivity of 0.907, specificity of 0.870, and accuracy of 0.893 in the training set; corresponding values in the testing set were 0.866 (95% CI: 0.767–0.966), 0.750, 0.800, and 0.769 (Fig. 4B, C; Table 4).The optimal cutoff value (threshold) for the nomogram was determined to be 0.320. At this specific threshold, the Decision Curve Analysis (DCA) indicated a net clinical benefit of 0.303 in the training cohort and 0.235 in the validation cohort. Decision Curve Analysis (DCA) further demonstrated favorable clinical utility across a range of threshold probabilities (Fig. 4D). Calibration performance, assessed using calibration curves (Supplementary Fig. 1), indicated good agreement between predicted and actual PNI probabilities. These results collectively indicate that the model achieves excellent predictive performance.
Fig. 4.
(A) Nomograms m to predict PNI in patients with ICC. (B) The area under receiver operating characteristic curve for the model in training cohort. (C) The area under receiver operating characteristic curve for the model in testing cohort. (D) The decision curve analysis (DCA) for the model
Table 4.
Performances of model for PNI prediction
| Cohort | AUC (95%CI) | Accuracy (%) | Sensitivity (%) | Specificity (%) |
|---|---|---|---|---|
| Train | 0.939 (0.891–0.988) | 0.893 | 0.907 | 0.870 |
| Test | 0.866 (0.767–0.966) | 0.769 | 0.750 | 0.800 |
Abbreviations: AUC, the area under the mean receiver operating characteristic curve
Discussion
PNI is crucial for prognostic evaluation and treatment strategy in clinical practice, critically influencing patient survival. In this study, we demonstrated that tumor location, target sign on DWI, satellite nodules, and peritumoral arterial hyper-enhancement were independent risk factors for PNI. The above four risk factors constituted the model for predicting PNI in patients with ICC. Following a thorough evaluation, the model was validated to possess robust and excellent predictive performance.Therefore, a preoperative model, mainly based on imaging features of Gd-DTPA-enhanced MRI, was able to excellently predict the PNI in patients with ICC. Our model, based on conventional MRI features, offers a practical alternative to more complex CT-based or radiomics models. Its reliance on visually assessed features enhances interpretability for clinicians and may facilitate easier integration into routine preoperative workflow without the need for specialized software, potentially improving multicenter reproducibility. The results of our study provides a simple and convenient tool for noninvasive prediction of PNI in ICC, which may be helpful to improve patient prognostic stratification and facilitate optimized and individualized treatment strategies.
The established prognostic factors for ICC (tumor size, number, vascular invasion, lymph node metastasis) within the AJCC staging system have limited predictive power [19]. Postoperative recurrence rates of 60–80% occur even in the absence of vascular invasion or lymph node metastasis, confirming the need for better prognostic tools [20, 21]. There is now compelling evidence that perineural invasion (PNI), a key prognostic feature in gastrointestinal cancers, serves as a critical marker in ICC, predicting poorer outcomes and increased recurrence [22, 23]. In the present study, the incidence rate of PNI was 38.2%, which was consistent with other research [11]. Nevertheless, current PNI diagnosis still relies on postoperative histopathology, revealing an urgent need for noninvasive preoperative prediction.
Our study identified four MRI features as independent predictors of PNI, which were collectively used to construct a preoperative prediction model. Notably, perihilar tumor location and the absence of a target sign on DWI align with the understanding of PNI as an aggressive, infiltrative phenotype. These features have also been linked to poor prognosis in other imaging studies of ICC [11, 15], supporting their biological relevance. The integration of these features into a simple, visually interpretable nomogram represents a distinct step toward clinical translation.
Interestingly, two of the model features—satellite nodules and peritumoral arterial hyper-enhancement—are peritumoral in nature. This may reflect an aggressive tendency of the tumor to invade beyond the capsule and infiltrate the surrounding parenchyma. Recent research has confirmed the crucial role of neuromodulation in shaping the immune microenvironment [24, 25]. Consistent with this concept, Meng et al. [26] demonstrated a significant association between PNI-positive ICC and a suppressed NK cell population alongside enhanced neutrophil infiltration. Our study identified peritumoral arterial hyper-enhancement as a critical feature for predicting PNI. A variety of studies have suggested that arterial peritumoral enhancement may be a result of compensatory arterial hyperperfusion leading to a reduced portal blood flow, which is associated with more aggressive biological behaviour and a poor prognosis [27, 28].This may help clarify whether PNI occurs more frequently in the tumor periphery and whether it correlates with specific immune microenvironment alterations and prognosis [29].
Tumor location was also confirmed as an independent predictor of PNI, with perihilar tumors accounting for approximately 80% of PNI-positive cases. Consistent with previous literature [30, 31], our results confirm that perihilar ICC is strongly associated with PNI. Putative mechanisms for this association include the frequently poor histologic differentiation and heightened invasiveness of perihilar ICC; alternatively, the abundant nerve plexus at the hepatic hilum may be a facilitating factor [32]. Target sign on DWI is a typical imaging feature of ICC based on Gd-DPTA-enhanced MRI, which is characterized by peripheral restricted diffusion and a less restricted central stroma [33].In our cohort, PNI-positive tumors more often showed marked DWI hyperintensity rather than a target sign, possibly reflecting a histologic pattern of high cellularity with sparse stroma, indicative of heightened biological activity and invasiveness. Consequently, the presence of a target sign was strongly associated with PNI-negative status (OR 0.03), reinforcing its value as an imaging marker in our model.
Although the model exhibited strong predictive performance, a slight decrease in accuracy was observed in the testing set compared to the training set. This may be attributable to sample size limitations and potential overfitting, which could affect generalizability. To mitigate overfitting, we employed a hold-out validation approach and confirmed the comparability of training and testing cohorts through equilibrium testing.
The study has several limitations. First, this was a retrospective, single-center study, which may have led to selection bias and limits the generalizability of our findings. While internal validation showed good performance, the risk of overfitting cannot be entirely ruled out. Therefore, external validation using independent, preferably multicenter, datasets is essential to confirm the model’s robustness and clinical applicability before widespread adoption. In the future research, we should aim to incorporate multicenter data and external validation, thereby improving the model’s applicability and robustness. In addition, further study should be performed to reveal the relationship between the imaging features and clinical features or genomic features, provide deeper insights into underlying pathophysiological processes.
Conclusion
In conclusion, we developed and validated a noninvasive preoperative model that integrates Gd-DTPA-enhanced MRI features to accurately predict PNI in ICC patients. This model shows significant potential to assist clinicians in preoperative risk assessment, thus facilitating improved prognostic stratification and the formulation of personalized treatment strategies.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
Not applicable.
Abbreviations
- PNI
Perineural invasion
- MRI
Magnetic resonance imaging
- ICC
Intrahepatic cholangiocarcinoma
- AUC
Area under receiver operating characteristic curve
- DCA
Decision curve analysis
Author contributions
JNY, PXP, ZSS conceived the project and designed the study. FYY, WYX, performed the data extraction and collection. ZSS performed the data analysis and wrote the manuscript. LYP, HD revised the manuscript. All authors contributed to the article and approved the submitted version.
Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Data availability
The data used in this study can be obtained by contacting the corresponding author.
Declarations
Ethics approval and consent to participate
This study was performed in accordance with the ethical standards of the Declaration of Helsinki. The protocol was approved by the Ethics Committee of the Eastern Hepatobiliary Surgery Hospital, the Third Affiliated Hospital of the Naval Medical University, China. For this retrospective study, the requirement for informed consent was waived by the Institutional Review Board (IRB).
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Sisi Zhang, Yayuan Feng and Da He contributed equally to this work and share first authorship.
Contributor Information
Ningyang Jia, Email: ningyangjia@163.com.
Xingpeng Pan, Email: 22590118@qq.com.
Yiping Liu, Email: liuyiping108@126.com.
References
- 1.Chen T, Zhang Y, Liu J, Rao Z, Wang M, Shen H, Zeng S. Trends in liver cancer mortality in China from 1990 to 2019: a systematic analysis based on the global burden of disease study 2019. BMJ Open. 2023;13(12):e074348. 10.1136/bmjopen-2023-074348. PMID: 38159955; PMCID: PMC10759138. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Cao G, Liu J, Liu M. Trends in mortality of liver disease due to hepatitis B in China from 1990 to 2019: findings from the global burden of disease study. Chin Med J (Engl). 2022;135(17):2049–55. PMID: 36228164; PMCID: PMC9746726. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.An L, Zheng R, Zhang S, Chen R, Wang S, Sun K, Lu L, Zhang X, Zhao H, Zeng H, Wei W, He J. Hepatocellular carcinoma and intrahepatic cholangiocarcinoma incidence between 2006 and 2015 in china: estimates based on data from 188 population-based cancer registries. Hepatobiliary Surg Nutr. 2023;12(1):45–55. 10.21037/hbsn-21-75. Epub 2021 Jul 21. PMID: 36860251; PMCID: PMC9944524. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Capuozzo M, Santorsola M, Ferrara F, et al. Intrahepatic cholangiocarcinoma biomarkers: towards early detection and personalized Pharmacological treatments. Mol Cell Probes. 2024;73:101951. 10.1016/j.mcp.2024.101951. [DOI] [PubMed] [Google Scholar]
- 5.Ribero D, Pinna AD, Guglielmi A, et al. Surgical approach for Long-term survival of patients with intrahepatic cholangiocarcinoma: A Multi-institutional analysis of 434 patients. Arch Surg. 2012;147(12):1107–13. 10.1001/archsurg.2012.1962. [DOI] [PubMed] [Google Scholar]
- 6.Liebig C, Ayala G, Wilks JA, Berger DH, Albo D. Perineural invasion in cancer: a review of the literature. Cancer. 2009;115(15):3379–91. 10.1002/cncr.24396. [DOI] [PubMed] [Google Scholar]
- 7.Shirai K, Ebata T, Oda K, et al. Perineural invasion is a prognostic factor in intrahepatic cholangiocarcinoma. World J Surg. 2008;32(11):2395–402. 10.1007/s00268-008-9726-2. [DOI] [PubMed] [Google Scholar]
- 8.Margies R, Gröger LK, Straub BK, Bartsch F, Lang H. Influence of Perineural (Pn), Lymphangio (L) and Vascular (V) Invasion on Survival after Resection of Perihilar Cholangiocarcinoma. Cancers (Basel). 2024;16(20):3463. Published 2024 Oct 12. 10.3390/cancers16203463 [DOI] [PMC free article] [PubMed]
- 9.Jiang JH, Fang DZ, Hu YT. Influence of surgical margin width on survival rate after resection of intrahepatic cholangiocarcinoma: a systematic review and meta-analysis. BMJ Open. 2023;13(5):e067222. 10.1136/bmjopen-2022-067222. Published 2023 May 8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Yu X, Zhang Z, Zou Y, et al. CT-based radiomics model for noninvasive prediction of progression-free survival in high-grade serous ovarian carcinoma: a multicenter study incorporating preoperative and postoperative clinical factors. BMC Med Imaging. 2025;25(1):320. 10.1186/s12880-025-01865-0. Published 2025 Aug 8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Liu Z, Luo C, Chen X, et al. Noninvasive prediction of perineural invasion in intrahepatic cholangiocarcinoma by clinicoradiological features and computed tomography radiomics based on interpretable machine learning: a multicenter cohort study. Int J Surg. 2024;110(2):1039–51. 10.1097/JS9.0000000000000881. Published 2024 Feb 1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Craven I, Griffiths PD, Hoggard N. Magnetic resonance imaging of epilepsy at 3 Tesla. Clin Radiol. 2011;66(3):278–86. 10.1016/j.crad.2010.10.010. [DOI] [PubMed] [Google Scholar]
- 13.Kovač JD, Janković A, Đikić-Rom A, Grubor N, Antić A, Dugalić V. Imaging spectrum of intrahepatic Mass-Forming cholangiocarcinoma and its mimickers: how to differentiate them using MRI. Curr Oncol. 2022;29(2):698–723. 10.3390/curroncol29020061. Published 2022 Jan 30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Tan G, Wang WQ, Yuan T, et al. Machine learning prediction of perineural invasion in intrahepatic cholangiocarcinoma. Eur J Surg Oncol. 2025;51(9):110203. 10.1016/j.ejso.2025.110203. [DOI] [PubMed] [Google Scholar]
- 15.Chen MC, Zhou XQ, Liu ZW, et al. Preoperative MRI prediction and molecular pathway study of perineural invasion in intrahepatic cholangiocarcinoma: insights from bioinformatics approach. Eur J Surg Oncol. 2026;52(1):110546. 10.1016/j.ejso.2025.110546. [DOI] [PubMed] [Google Scholar]
- 16.Qi Z, Yuan H, Li Q et al. An MRI-based fusion model for preoperative prediction of perineural invasion status in patients with intrahepatic cholangiocarcinoma. World J Surg Oncol. 2025;23(1):164. Published 2025 Apr 26. 10.1186/s12957-025-03819-w [DOI] [PMC free article] [PubMed]
- 17.Dunn M, Morgan MB, Beer TW. Perineural invasion: identification, significance, and a standardized definition. Dermatol Surg. 2009;35(2):214–21. 10.1111/j.1524-4725.2008.34412.x. [DOI] [PubMed] [Google Scholar]
- 18.Renzulli M, Brocchi S, Cucchetti A, et al. Can current preoperative imaging be used to detect microvascular invasion of hepatocellular carcinoma? Radiology. 2016;279(2):432–42. Epub 2015/12/15. 10.1148/radiol.2015150998 [DOI] [PubMed]
- 19.Kamarajah SK. Evaluation of the AJCC 8th edition staging system for pathologically versus clinically staged intrahepatic cholangiocarcinoma (iCCA): a time to revisit a dogma? A Surveillance, Epidemiology, and end results (SEER) analysis. J Gastrointest Cancer. 2019;50(3):392–9. 10.1007/s12029-018-0084-5. [DOI] [PubMed] [Google Scholar]
- 20.Yamamoto Y, Sugiura T, Okamura Y, et al. The evaluation of the eighth edition of the AJCC/UICC staging system for intrahepatic cholangiocarcinoma: a proposal of a modified new staging system. J Gastrointest Surg. 2020;24(4):786–95. 10.1007/s11605-019-04185-1. [DOI] [PubMed] [Google Scholar]
- 21.Kamarajah S, Giovinazzo F, Roberts KJ, et al. The role of down staging treatment in the management of locally advanced intrahepatic cholangiocarcinoma: review of literature and pooled analysis. Ann Hepatobiliary Pancreat Surg. 2020;24(1):6–16. 10.14701/ahbps.2020.24.1.6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Zhang Z, Zhou Y, Hu K, Wang D, Wang Z, Huang Y. Perineural invasion as a prognostic factor for intrahepatic cholangiocarcinoma after curative resection and a potential indication for postoperative chemotherapy: a retrospective cohort study. BMC Cancer. 2020;20(1):270. 10.1186/s12885-020-06781-w. Published 2020 Mar 30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Li Q, Chen C, Zhang J, et al. Prediction efficacy of prognostic nutritional index and Albumin-Bilirubin grade in patients with intrahepatic cholangiocarcinoma after radical resection: A Multi-Institutional analysis of 535 patients. Front Oncol. 2021;11:769696. 10.3389/fonc.2021.769696. Published 2021 Dec 10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Yang MW, Tao LY, Jiang YS, et al. Perineural invasion reprograms the immune microenvironment through cholinergic signaling in pancreatic ductal adenocarcinoma. Cancer Res. 2020;80(10):1991–2003. 10.1158/0008-5472.CAN-19-2689. [DOI] [PubMed] [Google Scholar]
- 25.Tan X, Sivakumar S, Bednarsch J, et al. Nerve fibers in the tumor microenvironment in neurotropic cancer-pancreatic cancer and cholangiocarcinoma. Oncogene. 2021;40(5):899–908. 10.1038/s41388-020-01578-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Meng XL, Lu JC, Zeng HY, et al. The clinical implications and molecular features of intrahepatic cholangiocarcinoma with perineural invasion. Hepatol Int. 2023;17(1):63–76. 10.1007/s12072-022-10445-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Teraoku H, Morine Y, Uyama N, et al. Role of central Hypo-enhancement in the hepatic arterial phase of dynamic computed tomography in patients with Mass-Forming intrahepatic cholangiocarcinoma. World J Surg. 2020;44(7):2350–8. 10.1007/s00268-020-05456-8. [DOI] [PubMed] [Google Scholar]
- 28.Cerrito L, Ainora ME, Borriello R, et al. Contrast-Enhanced imaging in the management of intrahepatic cholangiocarcinoma: state of Art and future perspectives. Cancers (Basel). 2023;15(13):3393. 10.3390/cancers15133393. Published 2023 Jun 28. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Panettieri E, Maki H, Kim BJ, et al. Arterial enhancement pattern predicts survival in patients with resectable and unresectable intrahepatic cholangiocarcinoma. Surg Oncol. 2022;40:101696. 10.1016/j.suronc.2021.101696. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Aishima S, Oda Y. Pathogenesis and classification of intrahepatic cholangiocarcinoma: different characters of Perihilar large duct type versus peripheral small duct type. J Hepatobiliary Pancreat Sci. 2015;22(2):94–100. 10.1002/jhbp.154. [DOI] [PubMed] [Google Scholar]
- 31.Aishima S, Kuroda Y, Nishihara Y, et al. Proposal of progression model for intrahepatic cholangiocarcinoma: clinicopathologic differences between hilar type and peripheral type. Am J Surg Pathol. 2007;31(7):1059–67. 10.1097/PAS.0b013e31802b34b6. [DOI] [PubMed] [Google Scholar]
- 32.Zeng D, Wang Y, Wen N, Li B, Cheng N, Lu J. Short-term and Long-term clinical outcomes of combined caudate lobectomy for intrahepatic cholangiocarcinoma involving the hepatic hilus: A propensity score analysis. Gut Liver. 2025;19(3):438–53. 10.5009/gnl240158. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Park HJ, Kim YK, Park MJ, Lee WJ. Small intrahepatic mass-forming cholangiocarcinoma: target sign on diffusion-weighted imaging for differentiation from hepatocellular carcinoma. Abdom Imaging. 2013;38(4):793–801. 10.1007/s00261-012-9943-x. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Citations
Supplementary Materials
Data Availability Statement
The data used in this study can be obtained by contacting the corresponding author.