Artificial intelligence and imaging for risk prediction of pancreatic cancer: a narrative review

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a lethal cancer accounting for over 90% of pancreatic cancer (PC) cases (1-3). PDAC is sporadic cancer that aggressively develops metastases. Although having a relatively low incidence rate, it is currently the 4^th leading cause of all cancer deaths in both males and females (1,4,5), with expectations to become the 2^nd most by 2030 (4,6,7). About 60,430 new cases and 48,220 deaths are expected to be PDAC-related in 2021 in the United States (8). Over 80% of PDAC diagnosis occurs at an advanced stage when complete surgical resection of tumor is complicated due to extensive vascular involvement and metastasis (9-12). This leaves a small portion of patients (<15%) having surgically resectable disease. As a result, the overall five-year survival rate (5YSR) of PDAC barely exceeds ~10%, though recent studies consistently suggest that early diagnosis of PDAC can improve treatment outcome and elevate the current 5YSR up to 50% (1,13,14). However, early-stage diagnosis is challenging as PDAC remains asymptomatic or presents nonspecific symptoms until cancer reaches advanced stage (3,11,15).

Predicting risk of PDAC allows subsequent targeted screening of high-risk individuals through longitudinal surveillance programs and consequently enhances cancer diagnosis at an early stage. Efficient risk prediction also enables treating certain precancerous conditions to decelerate or even prevent the development of PDAC. Unfortunately, risk prediction of PDAC is currently unresolved partially due to the absence of sensitive and specific biomarkers, lack of a viable prediction system, and low prevalence. Various approaches (16-18) have been proposed to identify individuals at high risk for PDAC or with early-stage disease by considering several clinical and epidemiologic characteristics including blood tumor markers, genetic biomarkers (e.g., familial PC), demographic characteristics, imaging findings (e.g., detectable lesions), pre-existing health conditions (e.g., pancreatitis, abdominal pain, recent onset diabetes), and lifestyle (e.g., weight loss, smoking, drinking alcohol) as risk indicators of PDAC. However, these indicators are nonspecific to PDAC and show association to a broad range of other diseases too. Thus, screening of a large population presenting these general indications is impractical, costly, and prone to result in false positives with the potential for leading to adverse outcomes considering risks of interventions to make the diagnosis and the anxiety in patients.

Studies show that the precancerous pancreas evolves through biological adaptations–expressed as morphological and textural variations on pancreatic imaging. Extensive analysis of such changes during or prior to development of PDAC via noninvasive abdominal imaging (e.g., MRI, CT, etc.) can provide insight to unique tissue characteristics and basis for precise prediction of PDAC. Systematic integration of these features with other indicators (e.g., clinical, genetic) can significantly improve estimating cancer risk and reduce prediction errors (e.g., false-positive rates), and unnecessary intensive examinations. However, such microlevel changes are imperceivable due to the complex location of pancreas, limited tissue contrast, and subjective bias, or neglected due to absence of connection to PDAC.

Artificial Intelligence (AI) offers enormous techniques for fast and extensive analysis of medical images for (I) accurate detection and classification of precancerous lesions, (II) discovering predictors through detailed examination of complex features, and (III) integration of multiple predicators to perform comprehensive risk assessment. AI models (19) for image analysis for risk prediction of several cancers have shown tremendous performance that surpasses traditional procedures in terms of accuracy and time efficiency. Unfortunately, the AI has not been yet fully utilized to address the challenges withholding efficient risk prediction of PDAC.

In this review article, the current risk prediction mechanisms for PDAC and their limitations are discussed. Supported by several peer examples of other cancers including breast cancer (BC), prostate cancer (PCa), and lung cancer (LC), a case is built as to how AI can efficiently assist in discovering highly specific predictors of PDAC through extensive assessment of pancreatic morphology and texture and precancerous lesions using abdominal imaging; followed by performing improved risk prediction by integrating the discovered predictors with conventional indicators of PDAC. The recent AI strategies to aid developing robust prediction models are also discussed. The purpose of this review is to emphasize the importance of abdominal imaging and AI application to resolve issues withholding efficient risk prediction of PDAC and provide readers with motivation and hope to expect improved management of PDAC in near future. We present the following article in accordance with the Narrative Review reporting checklist (available at https://cco.amegroups.com/article/view/10.21037/cco-21-117/rc).

Risk indicators of PDAC

A variety of complications and measurements are associated with the future incidence of PDAC. This includes clinical, hereditary, pre-existing health conditions, lifestyle, and demographic characteristics.

Elevations of certain blood compositions including carbohydrate antigen (CA)19-9 (20) and carcinoembryonic antigen (CEA) (21,22) are considered early indicators of PDAC. A recent study (23) claimed that CEMIP (cell migration-inducing hyaluronan binding protein), a newly identified protein, also called KIAA1199, can be a complementary marker to CA19-9 for enhanced prediction. However, the elevation could be due to other causes.

Also, genetic factors have been consistently linked with increased risk of developing PDAC. The familial pancreatic cancer (FPC) (24) describes that kindreds containing at least two first-degree relatives with PC have at ~10% higher risk to develop PDAC (24) than the general population. Genetic testing is usually the first assessment performed for FPC individuals to seek specific germline genetic mutations (25,26) including BRCA1, BRCA2, PALB2, hereditary pancreatitis, and Peutz-Jeghers syndrome.

Moreover, subjects with certain pre-existing health conditions are usually at high risk for PDAC. This includes pancreatic disorders such as pancreatitis that increases risk of PDAC by 2 to 3 times that of general population (27). Also, the new onset diabetes mellitus (NOD) has a bidirectional relationship with PDAC (28,29). Although its unknown when exactly NOD contributes to PDAC, there is established evidence that diabetes associated with ~25% of PDAC patients is diagnosed between 6 to 24 months before cancer diagnosis (30), suggesting that diabetes can both herald PDAC and act as a potential risk factor for PDAC. other studies have shown that subjects (>50 years) with NOD are at 6–8 times higher risk of developing PDAC within 3 years compared to non-diabetic population (28,29). Also, about 3–10% prevalence of PDAC is observed in NOD patients of age above 50 (31,32).

Furthermore, factors like smoking, alcohol (33), obesity (34), and poor diet (35) increases risk of PDAC. The risk in the FPC setting is observed even higher in smokers than in nonsmokers (36). The American Cancer Society has reported that chain cigarette smoking, and obesity increases the risk of PDAC by 25% and 20% respectively (37,38).

None of these factors is perfectly sufficient for predicting or confirming of PDAC as these factors are mostly general and are associated to a broad range of other diseases. Most of these do not even justify performing PDAC targeted screening. However, a combine assessment of all these factors may provide significant ‘clues’ to proceed with further screening or a good indication of future incidence of PDAC.

Image indicators

Pancreatic imaging plays a vital role in managing PDAC through prediction, diagnosis, staging, and prognosis. With several modalities including (e.g., CT, MRI) each with high versatility, pancreatic imaging provides a safe noninvasive means to look for specific potential predictors of PDAC. Several pancreatic complications occurred prior to or during the development of PDAC manifest unique morphological and textural changes on pancreas imaging. Such changes merit attention and can be regarded as credentialed predictors of PDAC. The distal parenchymal atrophy (39), intraductal papillary mucinous cancer of the pancreas (IPMNs) (40), and intraductal calculi (pancreatolithiasis) (41) are some of the disorders that cause progressive increase in tissue heterogeneity of the pancreas and are frequently considered as risk factors. In addition, pancreatic inflammation (42) and ductal dilatation (43) are consistently regarded as precancerous conditions that cause shape and size deformation of the pancreas and its components. Moreover, the signal intensity of the pancreatic region, where the tumor will likely develop, starts varying from peripheral regions during tumor development. For example, healthy cells turning into cancer start appearing darker on a CT scan. Furthermore, certain locations in the pancreas have relatively higher chances of developing tumor. For example, in 75–80% of the PDAC cases, the tumor develops in the head of the pancreas, while 20–25% chance for both body and tail (11,44). A consensus is that local changes in pancreatic subregions can provide concise measures to quantify risk. However, all these micro-level variations are difficult to determine by visual assessment, and therefore usually ignored or misperceived as normal, missing the opportunity to efficiently predict the risk of PDAC.

Risk prediction models of PDAC

Since clinical and genetic risk factors of PDAC lack sensitivity and specificity, imaging can be potential means to explore and appreciate adaptive features to cancer that can be integrated with existing indicators to achieve enhanced prediction. However, humans have limitations identifying, measuring, or interpreting such microlevel indicative features. Fortunately, the enormously advanced AI techniques for image processing and analysis together with conventional statistical methods offer insight into complex imaging data and reveal hidden patterns of information that assists in forecasting disease incidents.

Recently, conventional methods for risk prediction of several common cancers have been replaced by AI models that work by integrating multi-source risk factors (e.g., imaging, clinical) and perform enhanced risk prediction. For example, despite there are several risk prediction models (45-47) for BC including BCRAT (48), BCSC (49), BRCAPRO (50), BOADICEA (51), the Myriad model (52), the Rosner-Colditz model (53), the Tyrer-Cuzick (IBIS) model (54), and the Claus model (55), each utilizing variety of risk factors including hormonal and genetic indicators, radiation exposure, breast density, and lifestyle, the American Cancer Society guidelines urge to perform annual mammographic screening for women over 40 to have an enhanced risk prediction of BC. Since there is strong consensus that AI models could strengthen mammography analysis, several automated prediction models were recently proposed using imaging data alone that had surpassed the prediction accuracies of many of the existing non-image-based prediction models. Examples include a deep neural network (56,57), proposed in a multi-institutional study for assessment of breast density and beyond using mammograms, outperformed many existing risk prediction models. Also, a hybrid model (58) combining logistic regression and deep learning (DL) network, was trained on mammograms and outperformed Tyrer-Cuzick model and many others.

PCa is another example where AI has assisted to improve risk prediction. Recently, a hybrid model to stratify risk of PCa (59) was proposed that combined both machine learning (ML) and radiomic analysis of multiparametric MR images of the prostate and had achieved higher sensitivity and predictive value than conventional approaches. Also, a multi-institutional study (60) observed that ML techniques offer improved prostate specific antigen (PSA)-based risk stratification of PCa.

For LC, recently, promoting and implementing LC screening programs using low-dose CT imaging have resulted in improved diagnosis of LC at early-stage. To support such programs, a largescale study developed a deep ML model [referred to as DeepLR (61)] to predict LC risk using low-dose CT. The model is capable of accurately predicting which lesions will likely develop LC within a specific timeframe. The model showed excellent performance with a 1-, 2- and 3-year time-dependent AUC values for LC diagnosis of 0.968±0.013, 0.946±0.013, and 0.899±0.017.

AI for risk prediction of PDAC

Although AI applications remarkably improved risk prediction of many cancers, predicting PDAC using AI analysis of pancreatic imaging has remained forsaken, partially due to insufficient data to discover specific predictors in the precancerous pancreas, and perform model development and validation. Another possible reason is that imaging performed at the precancerous or early stage of PDAC usually comprises of CT examination which offers limited tissue discrimination to appreciate microlevel indicators of PDAC by naked eyes.

The peer examples indicate that AI can efficiently assist performing an in-depth examination of such images to discover and quantify predictors and systematically integrate such features with clinical and genetic indicators to perform enhanced prediction. To the best of our knowledge, only two research groups performed risk stratification of PDAC using CT images and AI.

A team at Kaiser Permanente Southern California (62) performed a retrospective study and developed a model that integrates known imaging (measured using CT images) and clinical risk factors to perform risk stratification of PDAC. They considered pancreas-related morphologic features from CT/MR images of individuals with their pancreas indicating ductal dilation (a potential PDAC predictor) at a precancerous stage. Some of the morphological and clinical features include atrophy, calcification, pancreatic cyst, and pancreatic ductal irregularity, and age, sex, race/ethnicity, tobacco, and alcohol use, respectively. Using a multi-state prediction model, they achieved reasonable discrimination (c-index 0.825–0.833) between those who developed PDAC and those who did not.

A team of imaging scientists and abdominal surgeons at Biomedical Imaging Research Institute of Cedars Sinai Medical Center (CSMC) Los Angeles extensively analyzed morphological and textural features of the pancreas in pre-diagnostic CT scans of PDAC patients (63). Their hypothesis was that the pre-diagnostic scans show unique features which are indicative of development of PDAC. They discovered several radiomic features which were significantly different than the normal and cancerous pancreas, and potentially predictive of PDAC. These scans were originally obtained for patients with non-PDAC reasons (e.g., other abdominal disorders due to vehicle accident or slip, etc.), who later (between 6 months to 3 years) developed PDAC. The discovered predictors were then used in a ML model to automatically perform risk stratification by classifying CT scans into control (healthy pancreas) and pre-diagnostic (high risk) classes. In the external validation process, their system achieved an average prediction accuracy of 86%. Deploying such a model in radiology room can significantly assist in identifying high risk individuals of PDAC. For example, it is reported that abdominal pain is the single most common reason for over 7 million patients visiting emergency room (ER) in the US; of which some subjects eventually develop PDAC in the next few years though remain undiagnosed until late stage. Since, CT examination is included in the current protocol of care for those having abdominal pain, a potential application to this proposed model would be to target such a group.

Both models perform reasonably well and justify the proposed design concept for risk prediction. However, the efficacy of these systems is still inadequate on account of insufficient data and model overfitting issues and thus require extensive external validation.

Discussion

Early diagnosis of PDAC is the call for immediate attention as the cases of PDAC are on the rise (10,64). The United States Prevention and Screening Task Force (USPSTF) has recommended a grade of “D” for screening for PDAC (65) in the general population suggesting that (I) imaging-based biomarkers for early detection are not yet established, and (II) the average risk population (~12 per 100,000) reduces the pre-test possibility of a laboratory test being truly positive. It is therefore the right time to take advantage of the emerging paradigm of AI and extensively explore the imaging features of pancreas to quantify the degree of elevated risk. With the support of AI, imaging of pancreas can be a vital source for classifying precancerous lesions, obtaining accurate measurements, discovering reliable predictors, and developing integrated systems to assist risk stratification of PDAC. The risk stratification of the PDAC will encourage follow-up targeted screening that could utilize AI-based techniques (66,67) to assist detecting PDAC in the early stage.

AI modelling for PDAC prediction (the opportunity)

In the recent years, the AI has addressed several potential challenges, related to deploying AI-based models, by offering alternate solutions.

Pretrained networks (68) offer a unique opportunity to train algorithms for tasks like lesion detection and classification, and pancreas segmentation even with a small amount of data. These AI networks are trained on thousands-to-millions of examples to perform similar tasks and can be adjusted to provide comparable performance for the current model.

Data privacy can be ensured when multicenter collaborative studies are performed to meet data requirements, maintain patients’ data privacy, and obtain multidisciplinary support during model development. The concept of federated learning (FL) (69) has made it possible to run AI algorithms for model training and validation without exposing or physically transferring the data to other sites. The FL schemes ensure the similar performance of the model as if all the data is residing on a single site without violating HIPPA and local institutional guidelines for data privacy.

Reference data labeling is a requirement for rigorous training and unbiased validation. Manual outlining of structures like pancreas, tumors, and lesions can be highly subjective, time-consuming, and prone to errors. The AI-based platforms provide mechanisms for radiologists from different sites to collaborate and ensure labeling consensus and produce highly reliable labels using commercial labeling software.

Computational efficiency is no longer a challenge as several institutes and commercial companies have started offering access to their hardware resources to run AI models with high-speed performance.

PDAC management—a future prospective

With the recent improvement in pre-operative, radiation, and chemotherapy regimens (by exploring new drugs and drug combinations), surgical innovations (such as irreversible electroporation (70,71), and strategies to enhance the immune response against PDAC, along with the current efforts to develop AI-imaging-based models for risk prediction, one can foresee significant improvement for PDAC management. Also, choosing the best imaging modality for PDAC screening within the multitude of options and when and how often imaging should be used in surveillance are still debatable issues.

In a nutshell, the fusion of AI and imaging can adequately address major issues complicating risk prediction of PDAC to a significant extent and offer better alternatives to current manual systems in terms of accuracy, time efficiency, and consistency. Undoubtedly, it is a potential research opportunity for scientists to further investigate and take advantage of state-of-the-art AI technologies to upgrade current prediction systems. Recently, the National Institute of Health at United States has started a large multicenter study at BIRI-CSMC to further enhance and validate their proposed risk prediction model on large datasets. With this strong mutual benefit for both the public and the health industry, a far superior PDAC management system is expected in near future.

Conclusions

This brief review emphasizes the importance of risk prediction of PDAC, and the role of AI and imaging to address issues related to risk prediction of PDAC. The application of AI-based imaging for risk prediction of three common cancers including BC, PCa, and LC was discussed. Moreover, the current challenges and recent AI advancement to overcome such complications were also discussed. A fair concern has been observed in health institutes and research supporting departments about early diagnosis of PDAC and the scope of risk prediction of PDAC. Efforts have been started to make progress and it can be expected that a better management of PDAC would be a reality soon.

Acknowledgments

Funding: This work was supported by the National Institute of Health (Grant No. 1 R01 CA260955-01) for the project “Predicting Pancreatic Ductal Adenocarcinoma Using Artificial Intelligence Analysis of Pre-diagnostic Computed Tomography Images”.

Footnote

Provenance and Peer Review: This article was commissioned by the Guest Editor (Savio George Barreto) for the series “Unresolved Issues in Pancreatic Cancer” published in Chinese Clinical Oncology. The article has undergone external peer review.

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://cco.amegroups.com/article/view/10.21037/cco-21-117/rc

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://cco.amegroups.com/article/view/10.21037/cco-21-117/coif). The series “Unresolved Issues in Pancreatic Cancer” was commissioned by the editorial office without any funding or sponsorship. The authors received grant funded by NIH (#: 1 R01 CA260955-01) for this study. The authors have no other 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.

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. Adamska A, Domenichini A, Falasca M. Pancreatic Ductal Adenocarcinoma: Current and Evolving Therapies. Int J Mol Sci 2017;18:1338. [Crossref] [PubMed]
2. Nolen BM, Brand RE, Prosser D, et al. Prediagnostic serum biomarkers as early detection tools for pancreatic cancer in a large prospective cohort study. PLoS One 2014;9:e94928. [Crossref] [PubMed]
3. Almeida PP, Cardoso CP, de Freitas LM. PDAC-ANN: an artificial neural network to predict pancreatic ductal adenocarcinoma based on gene expression. BMC Cancer 2020;20:82. [Crossref] [PubMed]
4. Muhammad W, Hart GR, Nartowt B, et al. Pancreatic Cancer Prediction Through an Artificial Neural Network. Front Artif Intell 2019;2:2. [Crossref] [PubMed]
5. Pannala R, Basu A, Petersen GM, et al. New-onset diabetes: a potential clue to the early diagnosis of pancreatic cancer. Lancet Oncol 2009;10:88-95. [Crossref] [PubMed]
6. Klein AP, Lindström S, Mendelsohn JB, et al. An absolute risk model to identify individuals at elevated risk for pancreatic cancer in the general population. PLoS One 2013;8:e72311. [Crossref] [PubMed]
7. Boursi B, Finkelman B, Giantonio BJ, et al. A Clinical Prediction Model to Assess Risk for Pancreatic Cancer Among Patients With New-Onset Diabetes. Gastroenterology 2017;152:840-850.e3. [Crossref] [PubMed]
8. American Cancer Society. Key Statistics for Pancreatic Cancer.
9. Kleeff J, Korc M, Apte M, et al. Pancreatic cancer. Nat Rev Dis Primers 2016;2:16022. [Crossref] [PubMed]
10. Orth M, Metzger P, Gerum S, et al. Pancreatic ductal adenocarcinoma: biological hallmarks, current status, and future perspectives of combined modality treatment approaches. Radiat Oncol 2019;14:141. [Crossref] [PubMed]
11. Stark A, Eibl G. Pancreatic ductal adenocarcinoma. Pancreapedia: The Exocrine Pancreas Knowledge Base. 2015 May 19.
12. Kaur S, Baine MJ, Jain M, et al. Early diagnosis of pancreatic cancer: challenges and new developments. Biomark Med 2012;6:597-612. [Crossref] [PubMed]
13. Ansari D, Aronsson L, Sasor A, et al. The role of quantitative mass spectrometry in the discovery of pancreatic cancer biomarkers for translational science. J Transl Med 2014;12:87. [Crossref] [PubMed]
14. Egawa S, Takeda K, Fukuyama S, et al. Clinicopathological aspects of small pancreatic cancer. Pancreas 2004;28:235-40. [Crossref] [PubMed]
15. Keane MG, Horsfall L, Rait G, et al. A case-control study comparing the incidence of early symptoms in pancreatic and biliary tract cancer. BMJ Open 2014;4:e005720. [Crossref] [PubMed]
16. Singhi AD, Koay EJ, Chari ST, et al. Early Detection of Pancreatic Cancer: Opportunities and Challenges. Gastroenterology 2019;156:2024-40. [Crossref] [PubMed]
17. Sharma A, Kandlakunta H, Nagpal SJS, et al. Model to Determine Risk of Pancreatic Cancer in Patients With New-Onset Diabetes. Gastroenterology 2018;155:730-739.e3. [Crossref] [PubMed]
18. Hart PA. Early Detection of Pancreatic Cancer in High-Risk Individuals: Where Do We Go From Here? Am J Gastroenterol 2019;114:560-1. [Crossref] [PubMed]
19. Kourou K, Exarchos TP, Exarchos KP, et al. Machine learning applications in cancer prognosis and prediction. Comput Struct Biotechnol J 2014;13:8-17. [Crossref] [PubMed]
20. Poruk KE, Gay DZ, Brown K, et al. The clinical utility of CA 19-9 in pancreatic adenocarcinoma: diagnostic and prognostic updates. Curr Mol Med 2013;13:340-51. [PubMed]
21. Meng Q, Shi S, Liang C, et al. Diagnostic and prognostic value of carcinoembryonic antigen in pancreatic cancer: a systematic review and meta-analysis. Onco Targets Ther 2017;10:4591-8. [Crossref] [PubMed]
22. van Manen L, Groen JV, Putter H, et al. Elevated CEA and CA19-9 serum levels independently predict advanced pancreatic cancer at diagnosis. Biomarkers 2020;25:186-93. [Crossref] [PubMed]
23. Lee HS, Jang CY, Kim SA, et al. Combined use of CEMIP and CA 19-9 enhances diagnostic accuracy for pancreatic cancer. Sci Rep 2018;8:3383. [Crossref] [PubMed]
24. Klein AP, Brune KA, Petersen GM, et al. Prospective risk of pancreatic cancer in familial pancreatic cancer kindreds. Cancer Res 2004;64:2634-8. [Crossref] [PubMed]
25. Carrera S, Sancho A, Azkona E, et al. Hereditary pancreatic cancer: related syndromes and clinical perspective. Hered Cancer Clin Pract 2017;15:9. [Crossref] [PubMed]
26. Axilbund JE, Wiley EA. Genetic testing by cancer site: pancreas. Cancer J 2012;18:350-4. [Crossref] [PubMed]
27. Kirkegård J, Mortensen FV, Cronin-Fenton D. Chronic Pancreatitis and Pancreatic Cancer Risk: A Systematic Review and Meta-analysis. Am J Gastroenterol 2017;112:1366-72. [Crossref] [PubMed]
28. Chari ST, Leibson CL, Rabe KG, et al. Probability of pancreatic cancer following diabetes: a population-based study. Gastroenterology 2005;129:504-11. [Crossref] [PubMed]
29. Sah RP, Nagpal SJ, Mukhopadhyay D, et al. New insights into pancreatic cancer-induced paraneoplastic diabetes. Nat Rev Gastroenterol Hepatol 2013;10:423-33. [Crossref] [PubMed]
30. Aggarwal G, Rabe KG, Petersen GM, et al. New-onset diabetes in pancreatic cancer: a study in the primary care setting. Pancreatology 2012;12:156-61. [Crossref] [PubMed]
31. Damiano J, Bordier L, Le Berre JP, et al. Should pancreas imaging be recommanded in patients over 50 years when diabetes is discovered because of acute symptoms? Diabetes Metab 2004;30:203-7. [Crossref] [PubMed]
32. Illés D, Terzin V, Holzinger G, et al. New-onset type 2 diabetes mellitus–a high-risk group suitable for the screening of pancreatic cancer? Pancreatology 2016;16:266-71. [Crossref] [PubMed]
33. Wang YT, Gou YW, Jin WW, et al. Association between alcohol intake and the risk of pancreatic cancer: a dose-response meta-analysis of cohort studies. BMC Cancer 2016;16:212. [Crossref] [PubMed]
34. Bracci PM. Obesity and pancreatic cancer: overview of epidemiologic evidence and biologic mechanisms. Mol Carcinog 2012;51:53-63. [Crossref] [PubMed]
35. Casari I, Falasca M. Diet and Pancreatic Cancer Prevention. Cancers (Basel) 2015;7:2309-17. [Crossref] [PubMed]
36. Petersen GM. Familial Pancreatic Adenocarcinoma. Hematol Oncol Clin North Am 2015;29:641-53. [Crossref] [PubMed]
37. Pancreatic Cancer Risk Factors. Available online: https://www.cancer.org/cancer/pancreatic-cancer/causes-risks-prevention/risk-factors.html
38. Pandol SJ, Apte MV, Wilson JS, et al. The burning question: why is smoking a risk factor for pancreatic cancer? Pancreatology 2012;12:344-9. [Crossref] [PubMed]
39. Ahn SS, Kim MJ, Choi JY, et al. Indicative findings of pancreatic cancer in prediagnostic CT. Eur Radiol 2009;19:2448-55. [Crossref] [PubMed]
40. Puckett Y, Sharma B, Kasi A. Intraductal Papillary Mucinous Cancer Of The Pancreas 2022; [PubMed]
41. Zhang GW, Zhou J, Li CL, et al. Diagnosis and treatment of pancreatolithiasis: report of 16 cases. Hepatobiliary Pancreat Dis Int 2005;4:135-8. [PubMed]
42. Shadhu K, Xi C. Inflammation and pancreatic cancer: An updated review. Saudi J Gastroenterol 2019;25:3-13. [Crossref] [PubMed]
43. Tanaka S, Nakaizumi A, Ioka T, et al. Main pancreatic duct dilatation: a sign of high risk for pancreatic cancer. Jpn J Clin Oncol 2002;32:407-11. [Crossref] [PubMed]
44. Elbanna KY, Jang HJ, Kim TK. Imaging diagnosis and staging of pancreatic ductal adenocarcinoma: a comprehensive review. Insights Imaging 2020;11:58. [Crossref] [PubMed]
45. Kim G, Bahl M. Assessing Risk of Breast Cancer: A Review of Risk Prediction Models. J Breast Imaging 2021;3:144-55. [Crossref] [PubMed]
46. Willoughby A, Andreassen PR, Toland AE. Genetic Testing to Guide Risk-Stratified Screens for Breast Cancer. J Pers Med 2019;9:15. [Crossref] [PubMed]
47. McCarthy AM, Guan Z, Welch M, et al. Performance of Breast Cancer Risk-Assessment Models in a Large Mammography Cohort. J Natl Cancer Inst 2020;112:489-97. [Crossref] [PubMed]
48. Costantino JP, Gail MH, Pee D, et al. Validation studies for models projecting the risk of invasive and total breast cancer incidence. J Natl Cancer Inst 1999;91:1541-8. [Crossref] [PubMed]
49. Tice JA, Bissell MCS, Miglioretti DL, et al. Validation of the breast cancer surveillance consortium model of breast cancer risk. Breast Cancer Res Treat 2019;175:519-23. [Crossref] [PubMed]
50. Berry DA, Iversen ES Jr, Gudbjartsson DF, et al. BRCAPRO validation, sensitivity of genetic testing of BRCA1/BRCA2, and prevalence of other breast cancer susceptibility genes. J Clin Oncol 2002;20:2701-12. [Crossref] [PubMed]
51. Lee A, Mavaddat N, Wilcox AN, et al. BOADICEA: a comprehensive breast cancer risk prediction model incorporating genetic and nongenetic risk factors. Genet Med 2019;21:1708-18. [Crossref] [PubMed]
52. Antoniou AC, Hardy R, Walker L, et al. Predicting the likelihood of carrying a BRCA1 or BRCA2 mutation: validation of BOADICEA, BRCAPRO, IBIS, Myriad and the Manchester scoring system using data from UK genetics clinics. J Med Genet 2008;45:425-31. [Crossref] [PubMed]
53. Rosner BA, Colditz GA, Hankinson SE, et al. Validation of Rosner-Colditz breast cancer incidence model using an independent data set, the California Teachers Study. Breast Cancer Res Treat 2013;142:187-202. [Crossref] [PubMed]
54. Warwick J, Birke H, Stone J, et al. Mammographic breast density refines Tyrer-Cuzick estimates of breast cancer risk in high-risk women: findings from the placebo arm of the International Breast Cancer Intervention Study I. Breast Cancer Res 2014;16:451. [Crossref] [PubMed]
55. McTiernan A, Kuniyuki A, Yasui Y, et al. Comparisons of two breast cancer risk estimates in women with a family history of breast cancer. Cancer Epidemiol Biomarkers Prev 2001;10:333-8. [PubMed]
56. Akselrod-Ballin A, Chorev M, Shoshan Y, et al. Predicting Breast Cancer by Applying Deep Learning to Linked Health Records and Mammograms. Radiology 2019;292:331-42. [Crossref] [PubMed]
57. Dembrower K, Liu Y, Azizpour H, et al. Comparison of a Deep Learning Risk Score and Standard Mammographic Density Score for Breast Cancer Risk Prediction. Radiology 2020;294:265-72. [Crossref] [PubMed]
58. Yala A, Lehman C, Schuster T, et al. A Deep Learning Mammography-based Model for Improved Breast Cancer Risk Prediction. Radiology 2019;292:60-6. [Crossref] [PubMed]
59. Toivonen J, Montoya Perez I, Movahedi P, et al. Radiomics and machine learning of multisequence multiparametric prostate MRI: Towards improved non-invasive prostate cancer characterization. PLoS One 2019;14:e0217702. [Crossref] [PubMed]
60. Perera M, Mirchandani R, Papa N, et al. PSA-based machine learning model improves prostate cancer risk stratification in a screening population. World J Urol 2021;39:1897-902. [Crossref] [PubMed]
61. Huang P, Lin CT, Li Y, et al. Prediction of lung cancer risk at follow-up screening with low-dose CT: a training and validation study of a deep learning method. Lancet Digit Health 2019;1:e353-62. [Crossref] [PubMed]
62. Chen W, Butler RK, Zhou Y, et al. Prediction of Pancreatic Cancer Based on Imaging Features in Patients With Duct Abnormalities. Pancreas 2020;49:413-9. [Crossref] [PubMed]
63. Qureshi TA, Gaddam S, Wachsman AM, et al. Predicting Pancreatic Ductal Adenocarcinoma Using Artificial Intelligence Analysis of Pre-diagnostic Computed Tomography Images. Journal of Cancer Biomarkers 2022; In Press. [Crossref] [PubMed]
64. da Costa WL Jr, Oluyomi AO, Thrift AP. Trends in the Incidence of Pancreatic Adenocarcinoma in All 50 United States Examined Through an Age-Period-Cohort Analysis. JNCI Cancer Spectr 2020;4:pkaa033.
65. HenriksonNBBowlesEJABlasiPR2019.
66. Chen PT, Chang D, Yen H, et al. Radiomic Features at CT Can Distinguish Pancreatic Cancer from Noncancerous Pancreas. Radiol Imaging Cancer 2021;3:e210010. [Crossref] [PubMed]
67. Liu KL, Wu T, Chen PT, et al. Deep learning to distinguish pancreatic cancer tissue from non-cancerous pancreatic tissue: a retrospective study with cross-racial external validation. Lancet Digit Health 2020;2:e303-13. [Crossref] [PubMed]
68. Kieffer B, Babaie M, Kalra S, et al. Convolutional neural networks for histopathology image classification: Training vs. using pre-trained networks. In 2017 Seventh International Conference on Image Processing Theory, Tools and Applications (IPTA) 2017 Nov 28 (pp. 1-6). IEEE.
69. Xu J, Glicksberg BS, Su C, et al. Federated Learning for Healthcare Informatics. J Healthc Inform Res 2020; [Epub ahead of print]. [PubMed]
70. Martin RC 2nd. Use of irreversible electroporation in unresectable pancreatic cancer. Hepatobiliary Surg Nutr 2015;4:211-5. [PubMed]
71. Ansari D, Kristoffersson S, Andersson R, et al. The role of irreversible electroporation (IRE) for locally advanced pancreatic cancer: a systematic review of safety and efficacy. Scand J Gastroenterol 2017;52:1165-71. [Crossref] [PubMed]