Nomogram based on preoperative contrast-enhanced CT predicts early liver metastasis in pancreatic ductal adenocarcinoma

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is the third leading cause of cancer-related death in the United States, the incidence of which is rising at a rate of 0.5–1.0% per year (1,2). Despite advances in neoadjuvant chemotherapy, interventional therapy, and surgical techniques for PDAC, the long-term prognosis of PDAC patients remains poor (3,4). An important reason for this is that 80% of PDAC patients have early liver metastases (LMs) (5). Synchronous LM originating from PDAC are commonly diagnosed at advanced disease stages, and the presence of LM is a strong predictor of poor outcomes (5-7). Therefore, early diagnosis of LM from PDAC is essential to prolong patient survival and develop individualized treatment plans.

At present, preoperative enhanced computed tomography (CT) for tumor, node, and metastasis (TNM) staging of patients with PDAC is recommended by the National Comprehensive Cancer Network (NCCN) (8). However, the ability of contrast-enhanced computed tomography (CECT) to detect LM <1 cm in size is limited and unsatisfactory, with an accuracy rate of only 50% (9-11). For CECT of suspected or uncertain PDAC with LM, magnetic resonance imaging (MRI) is another option (12); however, it is expensive, and some patients have contraindications, such as claustrophobia and metal implants, among others. In addition, MRI is not an essential tool for preoperative TNM staging of patients with PDAC. Previous studies have shown that radiomics and radiological features are associated with early LM of PDAC (13-15). However, the radiomics research process is complicated and thus not suited to clinical practice (16). A more practical method for routine clinical diagnosis would be a preoperative prediction model of LM from pancreatic cancer constructed using objective quantitative and qualitative data.

Therefore, this study aimed to construct a nomogram model to predict early LM from PDAC based on preoperative CECT and clinical features of patients with PDAC. We present this article in accordance with the CLEAR reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-24-2046/rc).

Methods

Ethical statement

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study was approved by the Institutional Review Board of Lanzhou University Second Hospital Medical Ethics Committee (NO. 2024A-414) and the requirement for individual consent for this analysis was waived due to the retrospective nature.

Patients

We retrospectively screened 224 patients with pathologically confirmed PDAC who visited Lanzhou University Second Hospital from January 2015 to November 2021. Early LM was defined as synchronous LM and metachronous LM (LM that occurred within 3 months after surgery). The inclusion criteria were as follows: synchronous and postoperative (≤3 months) LM confirmed by puncture pathology or typical imaging features; complete preoperative clinical data; complete preoperative CT scan data; and overall survival data. The exclusion criteria were as follows: history of preoperative tumor-related treatment; nondiagnostic CT image quality (i.e., motion- or metal-related artifacts); an interval of more than 1 month between the preoperative CT examination and pathologically confirmed PDAC; and the presence of other malignant tumors. Ultimately, 184 PDAC patients were included in the study. Figure 1 shows the patient recruitment flowchart. In this study, 184 patients were divided into training and validation sets according to the ratio of 6:4.

Figure 1 Flowchart of the training and validation cohorts. CT, computed tomography; LM, liver metastasis; NLM, non-early liver metastasis; PDAC, pancreatic ductal adenocarcinoma.

Clinical information and follow-up details

All clinical data were collected from the electronic medical records of patients, including age, sex, smoking status, alcoholism, chronic pancreatitis, diabetes, and tumor location (pancreas uncinate process, neck, or body tail). Laboratory analysis was performed of carcinoembryonic antigen (CEA; unit, ng/mL), carbohydrate antigen 125 (CA125; unit, U/mL), carbohydrate antigen 19-9 (CA19-9; unit, U/mL), alpha-fetoprotein (AFP; unit, ng/mL), plasma prothrombin time (PPT; unit, s), activated partial thromboplastin time (APTT; unit, s), thrombin time (TT; unit, s), and fibrinogen (FIB; unit, g/L). Clinical features included preoperative T and N staging of PDAC.

Preoperative CECT were defined as the CECT examination within 2 weeks before the patient underwent surgical treatment. Synchronous LM was defined as the presence of LMs on the first CECT examination. Metachronous LM is defined as LMs that occur within 3 months after PDAC surgery. For postoperative PDAC patients, it is recommended that follow-up appointments be scheduled every 3 months during the first year after surgery, every 3–6 months during years 2 to 3 post-operation, and subsequently every 6 months thereafter. The follow-up protocol includes serum tumor markers alongside abdominal ultrasound or contrast-enhanced CT scans. Some patients with PDAC were followed up for suspected LM by MRI scans. Extending the follow-up time or puncture pathology finally confirmed LM.

CT image acquisition

All CT examinations were performed using a 256-detector row spiral CT (Revolution CT; GE, Chicago, IL, USA) or a 128-detector row spiral CT scanner (Discovery 750HD CT; GE). Before the CT examination, each patient fasted for 6–8 hours. The CT scanning parameters were as follows: flat sweep tube voltage of 120 kVp, automatic milliampere second technology, tube current of 100–600 mA, collimator width of 0.625 mm, rack speed of 0.6 s/rot, pitch 0.983:1, and reconstruction layer thickness and layer spacing of 1.25 mm. During the CECT scans, iodixanol (320 mgI/mL) was injected through the anterior cubital vein with a high-pressure syringe, at a flow rate of 3.5–4.0 mL/s and dose of 1.0 mL/kg body weight. The trigger threshold for abdominal aorta monitoring was 50 Hounsfield units (HU). After the trigger, the arterial phase scan was 8 seconds, the venous phase scan was 30 seconds, and the delay phase scan was 120 seconds.

Image processing

Two radiologists blinded to the pathological and follow-up results analyzed each patient’s data. The experts reached a consensus regarding LM based on the typical imaging signs (17,18). In the event of disagreement in the evaluation of the characteristics for LM, a consensus was arrived at following consultation between the two radiologists. A region of interest (ROI) of approximately 40–60 mm² was drawn on the solid area of the PDAC on the preoperative CT image. The tumor density was determined on CT-enhanced three-phase images and expressed in HU. We observed the relationship between the tumor and blood vessels around the pancreas, and measured the maximum hug angle and minimum distance between the tumor and surrounding blood vessels (Figure 2).

Figure 2 Schematic diagram of imaging features of PDAC in CECT. (A-C) Schematic diagram of the measurement method for PDAC located in the uncinate process, neck, body, and tail. (D) Fuzzy diagram of the fat space around the tumor. (E) The shortest distance and maximum hug angle between the PDAC at the uncinate process site and the superior mesenteric artery/vein and peritumoral fat fuzzy of the tumor. (F) The shortest distance between the portal vein and tumor was 0, and the maximum hug angle was >180°. (G) The shortest distance between the splenic artery/vein and tumor was 0, and the maximum hug angle was >180°. At 3 months postoperatively, the patient developed LM. (H) The yellow circle indicates the fuzzy fat space around the tumor. CECT, contrast-enhanced computed tomography; LM, liver metastasis; PDAC, pancreatic ductal adenocarcinoma; PFS, peritumoral fat fuzzy.

The following quantitative characteristics were also analyzed: minimum distance between the tumor and superior mesenteric artery or splenic artery, defined as the TA distance; minimum distance between the tumor and superior mesenteric vein, portal vein, or splenic vein, defined as the TV distance; tumor size, expressed as the tumor longitudinal diameter (TLD) and tumor maximum transverse diameter (TMTD) of the lesion; and tumor density, expressed as HU, including the plain scan, arterial, portal, and delayed CT values in the tumor ROI. We used methods from previous studies to obtain the arterial, permeability, and perfusion indexes (1,14). The following quantitative characteristics were analyzed: maximum hug angle between the tumor and superior mesenteric artery or splenic artery, defined as the TA angle (>180° or ≤180°); maximum hug angle between the tumor and superior mesenteric vein, portal vein, or splenic vein, defined as the TV angle (>180° or ≤ 180°); and peritumoral fat fuzzy (PFS), defined as flocculant exudation in the area of normal fat density around the tumor.

Statistical analysis

We used R software (version 4.1.2; https://www.r-project.org/) for the statistical analysis. The threshold for statistical significance was set a priori at P<0.05. Continuous variables were expressed as mean ± standard deviation (SD) or median [interquartile range (IQR)]. Categorical variables were expressed as numbers and percentages. The LM and non-LM groups were compared using the Wilcoxon rank-sum test for continuous variables and λ² or Fisher’s exact test for categorical variables. We used the kappa test to evaluate the consistency of all CECT categorical feature extractions. The univariate logistic regression analyses included variables with statistically significant differences between the LM and non-early LM (NLM) groups. Multivariate logistic regression analysis was performed to identify independent predictors of LM, and included clinical (e.g., CEA and CA125) and radiological variables (e.g., TV angle and TV distance). Based on both the clinical and radiological features, we used the rms R package to build a nomogram model to predict LM from PDAC. Receiver operating characteristic (ROC), calibration, and decision curve analyses (DCA) were used to evaluate the model’s predictive performance, accuracy, and clinical applicability. Next, the samples were resampled 1,000 times using the bootstrap method for internal validation of the model, and a goodness-of-fit test was conducted to assess model stability.

Results

Patient characteristics

This study included 184 PDAC patients who were divided into an early LM group (n=74; mean age, 57.73±10.35 years; 51 males) and a NLM group (n=110; mean age, 57.55±9.08 years; 66 males) according to the time of the CECT examination. Table 1 summarizes the clinical baseline characteristics of the included patients.

We divided the 184 patients into training and validation sets with a ratio of 6:4. There were 110 patients in the training cohort (LM group, n=40; NLM group, n=70) and 73 in the validation cohort (LM group, n=34; NLM group, n=40). A total of 32 (17.39%) patients with PDAC underwent postoperative MRI follow-up. There were statistical differences in the preoperative CEA, CA125, T stage, N stage, and TT between the training and validation cohorts in the LM and NLM groups.

Feature extraction and comparison of CECT images

The two radiologists had high concordance (Kappa ≥0.753, P<0.001) in the readings of the CECT features of the 184 PDAC cases (Table 2). There were statistical differences between the LM and the NLM groups for TA distance, TV distance, maximum longitudinal diameter of the tumor, TA angle, TV distance, and PFS between the training and validation cohorts. The CECT features of the LM and NLM groups are shown in Table 3 and Figure 2.

Multivariable analysis and predictive nomogram

Based on the clinical characteristics and CT features shown in Table 1 and Table 3, we constructed a nomogram model of early LM from PDAC (Figure 3). Multivariate logistic regression analysis showed that TV distance [P=0.028, odds ratio (OR): 0.714, 95% confidence interval (CI): 0.514–0.948], TV angle >180° (P=0.017, OR: 6.718, 95% CI: 1.481–35.822), and PFS (P=0.025, OR: 5.893, 95% CI: 1.322–31.349) were independent risk factors for early LM in PDAC (Table 4).

Figure 3 Nomogram for predicting early LM from PDAC based on clinical and CECT features. The value assigned to each factor ranges from 0 to 100. The total score was obtained by adding up the scores for each factor. Based on the total score, the probability of early liver metastases from PDAC was illustrated by projecting the score to the bottom axis. “*”, factors demonstrating statistical significance in multivariate regression analysis. CA125, carbohydrate antigen 125; CA199, carbohydrate antigen 199; CEA, carcinoembryonic antigen; CECT, contrast-enhanced computed tomography; LM, liver metastasis; PDAC, pancreatic ductal adenocarcinoma; PFS, peritumoral fat fuzzy; TA angle, maximum hug angle between the tumor and superior mesenteric artery or splenic artery; TA distance, minimum distance between the tumor and superior mesenteric artery or splenic artery; TLD, tumor longitudinal diameter; TV angle, maximum hug angle between the tumor and superior mesenteric vein, portal vein, or splenic vein; TV distance, minimum distance between the tumor and superior mesenteric vein, portal vein, or splenic vein.

Model performance evaluation and validation

The diagnostic performance of the model for PDAC for early and non-early LM showed an area under the curve (AUC) of 0.944 (95% CI: 0.892–0.995) in the training cohort and an AUC of 0.921 (95% CI: 0.8577–0.985) in the validation cohort. The sensitivity and specificity of the training and validation cohorts were 82.5% (95% CI: 50.0–97.5%) and 94.3% (95% CI: 87.1–98.6%), and 82.4% (95% CI: 72.5–94.1%) and 92.5% (95% CI: 82.4–100%), respectively (Figure 4A,4B). In both the training and validation cohorts, the calibration curves showed that the model predicted the probability of early LM from PDAC and was in good agreement with the actual probability (goodness-of-fit test, P=0.389 and 0.699 >0.05; Figure 4C,4D). There was no statistical difference between the AUC values of the training and validation sets (P=0.387) (Figure 4E).

Figure 4 Nomogram-derived ROC curves for distinguishing early LMs of PDAC. (A,B) AUC of the nomogram model used to diagnose early LM in PDAC patients in the training cohort (A) and validation cohorts (B). (C,D) Calibration curves of combined models in the training cohort (C) and validation cohorts (D). In the calibration curve, the blue line indicates the model’s ideal prediction, and the red line reflects the real data distribution. (E) Comparison of AUC values between training (blue line) and validation (red line) sets. AUC, area under the curve; CECT, contrast-enhanced computed tomography; CI, confidence interval; FPR, false positive rate; LM, liver metastasis; PDAC, pancreatic ductal adenocarcinoma; ROC, receiver operating characteristic; TPR, true positive rate.

Subgroup analysis and clinical decision curve of early LM from PDAC

For LMs smaller than 10 mm, ROC-AUC, sensitivity, and specificity of the model for predicting PDAC LM were 0.972 (95% CI: 0.937–1.000), 97.20% (95% CI: 85.5–99.9%), and 90.00% (95% CI: 80.5–95.9%), respectively. The clinical decision curve was constructed from the training cohort data, and the results showed that under the threshold probability of 10%, the model that we constructed obtained the maximum net benefit (Figure 5).

Figure 5 Nomogram clinical decision curve. The nomogram model achieved the maximum net benefit with a threshold probability of 10%.

Discussion

Early assessment of LM from PDAC can aid patient management and help to avoid unnecessary surgery. However, preoperative CECT can easily miss small lesions associated with early LM in PDAC (12,19,20). This study analyzed preoperative clinical and CECT features of 184 PDAC patients, and found that some CECT features showed good diagnostic performance for early LM from PDAC. Furthermore, these features were not limited to patients with LM found on CECT for the first time (1,21). Multivariate analysis found that TV distance, TV angle >180°, and PFS were independent risk factors for early LM from PDAC.

We observed that preoperative CEA and CA125 tumor marker levels were significantly higher in patients with early LM from PDAC than NLM patients. As previously reported, CEA and CA125 can predict LM from PDAC, where elevated serum CA125 concentrations may represent highly aggressive tumors and early micro-metastases (22,23). In addition, some studies have found that the serum concentration of CA19-9, N-stage, and T-stage can also predict early LM from PDAC (24-26). CA19-9 and TN-stage were not statistically different between the early and non-early LM groups in our study. This may be because we included patients with LM within 3 months after surgery, which reduced the differences in preoperative serological tumor markers and TNM stage. Secondly, the limited sample size resulted in CA19-9 and TN-stage failing to demonstrate statistical significance in the analysis. Existing research has established CA19-9 as an independent predictor of occult LM (27-29). Therefore, we maintain that CA19-9 persists as a significant predictor for LM, despite the absence of statistically significant differentiation in this specific study cohort.

Quantitative analysis of preoperative CECT images showed that TA distance, TV distance, TA angle, TV angle, TLD, and PFS were statistically significantly different between early and non-early LM. Rigiroli et al. found that the maximum hug angle and minimum distance between the tumor and superior mesenteric artery were independent predictors of PDAC invasion of the superior mesenteric artery (30). The radiomics model constructed from CECT scans had good predictive performance for PDAC superior mesenteric artery invasion (AUC =0.71; 95% CI: 0.62–0.79). Also, the angle (>180°) and minimum distance between the tumor and portal vein, splenic vein, superior mesenteric vein, superior mesenteric artery, and splenic artery could predict the occurrence of early LM from PDAC. This may be because tumor cells invaded the above veins and entered the liver through the portal vein to form early intrahepatic micro-metastases (1,30,31). D’Onofrio et al. found that the arterial- and permeability indexes could predict LM from PDAC (14). However, we only found statistical differences in the arterial and permeability indexes between the LM and the NLM groups in the validation cohort. Furthermore, the CECT features showed a statistically significant difference between early and non-early LM PDAC patients with PFS. Due to the perineural growth pattern of PDAC, tumor cells will infiltrate the surrounding tissue structure, which appears as blurred fat space around the tumor on CECT (32-34). This also suggests that the tumor cells may have invaded the surrounding blood vessels and nerves.

Through univariate analysis of preoperative clinical factors and CECT features in PDAC patients, CEA, CA125, TA distance, TV distance, TA angle, TV angle, TLD, and PFS were included in multivariate regression analyses to construct the nomogram model. The AUC, sensitivity, and specificity of the training and validation cohorts based on the nomogram model were 0.944, 95.0%, and 88.6%, and 0.972, 97.1%, and 85.0%, respectively. Zhang et al. performed a predictive analysis of PDAC LM using a radiomics model and showed that it had good discriminative performance in the training (AUC =0.93) and validation cohorts (AUC =0.81) (15). Hang et al. performed CT texture analysis of PDAC and found that radiomics scores could predict overall survival in patients with LM from PDAC (13). Thus, radiomics can aid the prediction of early LM in PDAC patients. However, the radiomics research process is complicated and thus not suited to clinical practice (35). The model developed in this study demonstrates superior clinical practicality and enhanced applicability compared to conventional radiomics approaches. Furthermore, the radiological features incorporated in this framework exhibit strong inter-observer consistency and robust stability, demonstrating reduced susceptibility to subjective evaluation heterogeneity. Notably, this predictive tool offers improved clinical interpretability when contrasted with current image-based machine learning models for early detection of PDAC LM.

This study had several limitations. First, it was a single-center retrospective study lacking validation with external datasets. However, our internal validation cohort demonstrated the discriminative performance and robustness of the model. In the future, we will pursue multi-center collaborations to further validate the model’s generalization capability. Second, for most patients, synchronous LM (51/74, 68.9%) could be detected by the first CECT. At the same time, the results of the subgroup analysis were subject to overfitting due to the insufficient data of cases with LM smaller than 10 mm in diameter. Therefore, when the LM of PDAC is less than 10 mm, the prediction efficacy of the model cannot be effectively estimated. The predictions of metachronous LM within 3 months after PDAC surgery may have been biased and require validation with a larger dataset. Finally, MRI can improve the detection rate of early LM in PDAC patients, and our patients lacked preoperative MRI data for comparative analysis (12). In our future research, we will further develop a prospective controlled trial to evaluate the clinical utility of MRI in detecting PDAC-associated LM.

Conclusions

Among the features of preoperative CECT, TV distance, TV angle >180°, and PFS were independent predictors of early LM from PDAC. The nomogram can predict early LM from PDAC, thereby assisting clinical treatment decision-making.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the CLEAR reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-24-2046/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-24-2046/dss

Funding: This study was supported by grants from the Natural Science Foundation of Gansu Province (Nos. 24JRRA332 and 25JRRA774) and the General Project of Cuiying Student Research Cultivation Program (No. CYXZPT2025-06).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-24-2046/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Helsinki Declaration and its subsequent amendments. This retrospective study was approved by the institutional review board of Lanzhou University Second Hospital Medical Ethics Committee (No. 2022A-298) and individual consent for this analysis was waived due to the retrospective nature.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Park W, Chawla A, O'Reilly EM. Pancreatic Cancer: A Review. JAMA 2021;326:851-62. [Crossref] [PubMed]
2. Qian ZR, Rubinson DA, Nowak JA, Morales-Oyarvide V, Dunne RF, Kozak MM, et al. Association of Alterations in Main Driver Genes With Outcomes of Patients With Resected Pancreatic Ductal Adenocarcinoma. JAMA Oncol 2018;4:e173420. [Crossref] [PubMed]
3. Wagner M, Redaelli C, Lietz M, Seiler CA, Friess H, Büchler MW. Curative resection is the single most important factor determining outcome in patients with pancreatic adenocarcinoma. Br J Surg 2004;91:586-94. [Crossref] [PubMed]
4. Ryan DP, Hong TS, Bardeesy N. Pancreatic adenocarcinoma. N Engl J Med 2014;371:1039-49. [Crossref] [PubMed]
5. Tsilimigras DI, Brodt P, Clavien PA, Muschel RJ, D'Angelica MI, Endo I, Parks RW, Doyle M, de Santibañes E, Pawlik TM. Liver metastases. Nat Rev Dis Primers 2021;7:27. [Crossref] [PubMed]
6. Conroy T, Desseigne F, Ychou M, Bouché O, Guimbaud R, Bécouarn Y, et al. FOLFIRINOX versus gemcitabine for metastatic pancreatic cancer. N Engl J Med 2011;364:1817-25. [Crossref] [PubMed]
7. Tempero MA. NCCN Guidelines Updates: Pancreatic Cancer. J Natl Compr Canc Netw 2019;17:603-5. [Crossref] [PubMed]
8. National Comprehensive Cancer Network. Pancreatic Adenocarcinoma: NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines). 2021 (Version 2). [EB/OL]. https://www.nccn.org.
9. Niekel MC, Bipat S, Stoker J. Diagnostic imaging of colorectal liver metastases with CT, MR imaging, FDG PET, and/or FDG PET/CT: a meta-analysis of prospective studies including patients who have not previously undergone treatment. Radiology 2010;257:674-84.
10. Kuszyk BS, Bluemke DA, Urban BA, Choti MA, Hruban RH, Sitzmann JV, Fishman EK. Portal-phase contrast-enhanced helical CT for the detection of malignant hepatic tumors: sensitivity based on comparison with intraoperative and pathologic findings. AJR Am J Roentgenol 1996;166:91-5. [Crossref] [PubMed]
11. Tsurusaki M, Numoto I, Oda T, Wakana M, Suzuki A, Yagyu Y, Matsuki M, Ishii K. Assessment of Liver Metastases Using CT and MRI Scans in Patients with Pancreatic Ductal Adenocarcinoma: Effects of Observer Experience on Diagnostic Accuracy. Cancers (Basel) 2020;12:1455. [Crossref] [PubMed]
12. Gassert FG, Ziegelmayer S, Luitjens J, Gassert FT, Tollens F, Rink J, Makowski MR, Rübenthaler J, Froelich MF. Additional MRI for initial M-staging in pancreatic cancer: a cost-effectiveness analysis. Eur Radiol 2022;32:2448-56. [Crossref] [PubMed]
13. Hang J, Xu K, Yin R, Shao Y, Liu M, Shi H, Wang X, Wu L. Role of CT texture features for predicting outcome of pancreatic cancer patients with liver metastases. J Cancer 2021;12:2351-8. [Crossref] [PubMed]
14. D'Onofrio M, De Robertis R, Aluffi G, Cadore C, Beleù A, Cardobi N, Malleo G, Manfrin E, Bassi C. CT Simplified Radiomic Approach to Assess the Metastatic Ductal Adenocarcinoma of the Pancreas. Cancers (Basel) 2021;13:1843. [Crossref] [PubMed]
15. Zhang T, Dong X, Zhou Y, Liu M, Hang J, Wu L. Development and validation of a radiomics nomogram to discriminate advanced pancreatic cancer with liver metastases or other metastatic patterns. Cancer Biomark 2021;32:541-50. [Crossref] [PubMed]
16. Mayerhoefer ME, Materka A, Langs G, Häggström I, Szczypiński P, Gibbs P, Cook G. Introduction to Radiomics. J Nucl Med 2020;61:488-95. [Crossref] [PubMed]
17. Wernecke K, Peters PE. Sonographic and computed tomographic diagnosis of liver metastases. A review. Radiologe 1985;25:141-51.
18. Danet IM, Semelka RC, Nagase LL, Woosely JT, Leonardou P, Armao D. Liver metastases from pancreatic adenocarcinoma: MR imaging characteristics. J Magn Reson Imaging 2003;18:181-8. [Crossref] [PubMed]
19. Zhang Y, Huang J, Chen M, Jiao LR. Preoperative vascular evaluation with computed tomography and magnetic resonance imaging for pancreatic cancer: a meta-analysis. Pancreatology 2012;12:227-33. [Crossref] [PubMed]
20. Motosugi U, Ichikawa T, Morisaka H, Sou H, Muhi A, Kimura K, Sano K, Araki T. Detection of pancreatic carcinoma and liver metastases with gadoxetic acid-enhanced MR imaging: comparison with contrast-enhanced multi-detector row CT. Radiology 2011;260:446-53. [Crossref] [PubMed]
21. Guan Z, Li ZW, Yang D, Yu T, Jiang HJ, Zhang XY, Yan S, Hou W, Sun YS. Small arteriole sign: an imaging feature for staging T4a colon cancer. Eur Radiol 2024;34:444-54. [Crossref] [PubMed]
22. Liu W, Tang B, Wang F, Qu C, Hu H, Zhuang Y, Gao H, Xie X, Tian X, Yang Y. Predicting early recurrence for resected pancreatic ductal adenocarcinoma: a multicenter retrospective study in China. Am J Cancer Res 2021;11:3055-69.
23. He C, Huang X, Zhang Y, Cai Z, Lin X, Li S. A Quantitative Clinicopathological Signature for Predicting Recurrence Risk of Pancreatic Ductal Adenocarcinoma After Radical Resection. Front Oncol 2019;9:1197. [Crossref] [PubMed]
24. Nakagawa K, Akahori T, Nishiwada S, Nagai M, Nakamura K, Tanaka T, Tamamoto T, Ohbayashi C, Hasegawa M, Kichikawa K, Ikeda N, Sho M. Prognostic factors for actual long-term survival in the era of multidisciplinary treatment for pancreatic ductal adenocarcinoma. Langenbecks Arch Surg 2018;403:693-700. [Crossref] [PubMed]
25. Takagi C, Kikuchi Y, Shirakawa H, Hoshimoto S, Tomikawa M, Ozawa I, Hishinuma S, Ogata Y. Predictive Factors for Elevated Postoperative Carbohydrate Antigen 19-9 Levels in Patients With Resected Pancreatic Cancer. Anticancer Res 2019;39:3177-83. [Crossref] [PubMed]
26. Ji F, Fu SJ, Guo ZY, Pang H, Ju WQ, Wang DP, Hua YP, He XS. Prognostic value of combined preoperative lactate dehydrogenase and alkaline phosphatase levels in patients with resectable pancreatic ductal adenocarcinoma. Medicine (Baltimore) 2016;95:e4065. [Crossref] [PubMed]
27. Charbonneau J, Noël P, Ouellet JF, Ouellet JB, Deshaies I, Daigle C, Ouellet É. Brind'Amour A. Diagnostic Value of Diffusion-Weighted MRI for the Detection of Peritoneal and Liver Metastases in High-Risk Pancreatic Ductal Adenocarcinoma: A Prospective Pilot Study. Ann Surg Oncol 2025;32:6511-8. [Crossref] [PubMed]
28. Wang F, Xu X, Chen Y, Xu J, Ji W, Yu D. Evaluation of Occult Liver Metastases in Pancreatic Adenocarcinoma by Diffusion-Weighted Related Magnetic Resonance Imaging. J Magn Reson Imaging 2025;62:536-48. [Crossref] [PubMed]
29. Yee EJ, Torphy RJ, Thielen ON, Easwaran L, Franklin O, Sugawara T, Bartsch C, Garduno N, McCarter MM, Ahrendt SA, Schulick RD, Del Chiaro M. Radiologic Occult Metastases in Pancreatic Cancer: Analysis of Risk Factors and Survival Outcomes in the Age of Contemporary Neoadjuvant Multi-agent Chemotherapy. Ann Surg Oncol 2024;31:6127-37. [Crossref] [PubMed]
30. Rigiroli F, Hoye J, Lerebours R, Lafata KJ, Li C, Meyer M, Lyu P, Ding Y, Schwartz FR, Mettu NB, Zani S Jr, Luo S, Morgan DE, Samei E, Marin D. CT Radiomic Features of Superior Mesenteric Artery Involvement in Pancreatic Ductal Adenocarcinoma: A Pilot Study. Radiology 2021;301:610-22. [Crossref] [PubMed]
31. Chang ST, Jeffrey RB, Patel BN, DiMaio MA, Rosenberg J, Willmann JK, Olcott EW. Preoperative Multidetector CT Diagnosis of Extrapancreatic Perineural or Duodenal Invasion Is Associated with Reduced Postoperative Survival after Pancreaticoduodenectomy for Pancreatic Adenocarcinoma: Preliminary Experience and Implications for Patient Care. Radiology 2016;281:816-25. [Crossref] [PubMed]
32. Tranchart H, Gaujoux S, Rebours V, Vullierme MP, Dokmak S, Levy P, Couvelard A, Belghiti J, Sauvanet A. Preoperative CT scan helps to predict the occurrence of severe pancreatic fistula after pancreaticoduodenectomy. Ann Surg 2012;256:139-45. [Crossref] [PubMed]
33. Low G, Panu A, Millo N, Leen E. Multimodality imaging of neoplastic and nonneoplastic solid lesions of the pancreas. Radiographics 2011;31:993-1015. [Crossref] [PubMed]
34. Steiger P. Radiomics and Artificial Intelligence: From Academia to Clinical Practice. Radiology 2022;303:542-3. [Crossref] [PubMed]
35. Janssen BV, Verhoef S, Wesdorp NJ, Huiskens J, de Boer OJ, Marquering H, Stoker J, Kazemier G, Besselink MG. Imaging-based Machine-learning Models to Predict Clinical Outcomes and Identify Biomarkers in Pancreatic Cancer: A Scoping Review. Ann Surg 2022;275:560-7. [Crossref] [PubMed]