Introduction

Early detection of gastric cancer may result from adequate diagnosis and surveillance of individuals with gastric intestinal metaplasia (GIM). Computer-assisted diagnosis systems based on artificial intelligence, or more specifically Deep Neural Networks (DNN) applied to esophagogastroduodenoscopy may be crucial in optimizing and standardizing patients’ final diagnosis. Literature has shown promising DNN results for single datasets, some of which reaching the 90% accuracy suggested by the European Society of Gastrointestinal Endoscopy for these algorithms [1]. Moreover, traditionally Wight-Light Endoscopy is referred as suboptimal to determine the presence of GIM. We aimed to determine the accuracy of current state-of-the-art AI algorithms for WLE from 2 different scopes’ providers for GIM detection.

Aims & Methods

Two datasets were used in these experiments:
• Dataset A (Fujifilm) was used for training and testing, consisting of a private benchmark dataset [1] of which we have used 4876 endoscopic White Light (WL) images (2438 normal, 2438 GIM).
• Dataset B (Olympus) was used for external validation and was collected from a lower quality clinical routine dataset, including 373 WL images (170 normal, 203 GIM). All images were annotated by clinical experts and the primary outcome was GIM classification accuracy.

We conducted experiments comparing the performance of DNNs (ResNet50) in GIM detection by training the models with and without the integration of synthetic/artificially generated endoscopy images. This involved employing four different methods: using the default dataset, augmenting the dataset with classic data augmentation transformations, incorporating images generated by Deep Convolutional Generative Adversarial Networks (DCGAN), and integrating images generated by Wasserstein GAN (WGAN). A part of dataset A (15%) was saved for testing in the different experiments and the remaining data (85%) was used to train the model, using a 5-fold cross-validation. In the end, dataset B was used as an external test set to evaluate the generalization capabilities of the different DNNs.

Results

The ResNet50 (dataset A – test set) achieved an accuracy of 92% for the default dataset, 93% for the data augmentation transformations, 93% for the DCGAN and 94% for the WGAN. The respective sensitivity values were 92%, 91%, 92% and 92%. Then, these models were evaluated in the dataset B, to infer its generalization capabilities. The ResNet50 achieved an accuracy of 63% for the default dataset, 68% for the data augmentation transformations, 65% for the DCGAN and 65% for the WGAN. The respective sensitivity values were 66%, 80%, 73% and 73%.

	Single Dataset (A->A) [Fujifilm WLE)				Generalization (A->B) [Olympus WLE]
ResNet50	Default	Augm.	DCGAN	WGAN	Default	Augm.	DCGAN	WGAN
Accuracy	0.92	0.93	0.93	0.94	0.63	0.68	0.65	0.65
Sensitivity	0.92	0.91	0.92	0.92	0.66	0.80	0.73	0.70

Conclusion

​Even for WLE, current DNN architectures seem to obtain very good results for GIM detection. However, generalization is still insufficient, since validation in different datasets (different clinical settings, different endoscopic technologies) obtains average detection accuracy and sensitivity, illustrating the excessive specialization caused by maximizing accuracy results in single datasets.

References

1] Messmann H, Bisschops R, Antonelli G, Libânio D, Sinonquel P, Abdelrahim M, Ahmad OF, Areia M, Bergman JJGHM, Bhandari P, Boskoski I, Dekker E, Domagk D, Ebigbo A, Eelbode T, Eliakim R, Häfner M, Haidry RJ, Jover R, Kaminski MF, Kuvaev R, Mori Y, Palazzo M, Repici A, Rondonotti E, Rutter MD, Saito Y, Sharma P, Spada C, Spadaccini M, Veitch A, Gralnek IM, Hassan C, Dinis-Ribeiro M. Expected value of artificial intelligence in gastrointestinal endoscopy: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy. 2022 Dec;54(12):1211-1231. doi: 10.1055/a-1950-5694. Epub 2022 Oct 21. PMID: 36270318.

[2] Yang J, Ou Y, C hen Z, Liao J, Sun W, Luo Y, et al. A Benchmark Dataset of Endoscopic Images and Novel Deep Learning Method to Detect Intestinal Metaplasia and Gastritis Atrophy. IEEE J Biomed Health Inform 2022:1 10., https://doi.org/10.1109/JBHI.2022.3217944.

Introduction

Helicobacter pylori (H. pylori) is a bacterium which inhabits the gastric epithelium and has been associated as a causative agent for dyspepsia, chronic gastritis, peptic and duodenal ulceration and gastric cancer. Testing for this pathogen is by Urea Breath Test (UBT) or OGD, and treatment varies according to local sensitivities and guidelines. H. pylori rates in developed countries have been falling in recent years as have those of associated conditions including gastric cancer, however increases in antimicrobial resistance recently has slowed this trend.
The current burden of disease either in asymptomatic population or symptomatic in Ireland is unknown, as are the rates of reinfection and recrudescence. Having up-to-date data on local epidemiology could have an effect on our strategy to detect and manage this class 1 carcinogen..

Aims & Methods

All UBTs performed in a tertiary hospital since 2016 were retrospectively assessed.
Demographic data & test results were taken from the UBT computer database and electronic patient record. All patients returning positive tests were offered appropriate antimicrobial therapy based on up to date local guidelines, and follow-up post-eradication testing by repeat UBT.

Results

From 2016-2023, 6716 UBTs were performed. 32 were excluded due to incomplete data. N=6684. 2752 (41.16%) were male, mean age was 43. Overall, 1589 (23.77%) tests were positive. There was no significant difference in positivity by gender (p=0.321). However there was a slight positive correlation with increasing age (AUC=0.5178 p=0.035). The positivity rate in our cohort reduced by roughly 2% per annum from 2016-22 (30%-18.5%).
Data on UBT type (Diagnostic vs Post-Eradication) was available from 2020-2023 when 3496 tests were performed and 2570 (73.5%) were diagnostic tests: 1060 (41.2%) male, mean age 41 years, 48 (19%) were positive. Positivity between genders was similar (p=0.3771). Conversely, in the post-eradication group n=926 (412, 44.5% Male), 259 (28%) were positive, females were more likely to remain positive post-therapy (32% vs 23%, X2=8.031, p=0.0046).
Over the 8 years, 747 patients returned positive tests (306 (41%) male). All were offered follow up testing. Males were significantly less likely to attend follow-up testing (140/306, 46% vs 254/441, 58%, p=0.0017)
In all 83 (2.4%) patients were re-tested for H. pylori after returning negative UBT tests due to persistent symptoms. Mean time to re-testing was 270 days, 34 (40%) were positive, representing 0.97% of the cohort.. Both genders were equally likely to have had a false negative UBT, male 10/30 versus 24/77 women, p=0.8214)

Conclusion

The overall positivity rate for diagnostic tests over the past 8 years was (19%) and is falling year on year.
The high positivity rate for post-eradication tests (28%) reflects rising antibiotic resistance.
The overall attendance rates for post-eradication tests after positive results were disappointing
Finally, the false negative rate in patients with persistent symptoms was high, 40% retested remained positive, raising the possibility of residual infection in our asymptomatic cohort.

References

1. Chen, Y.C., et al., Global Prevalence of Helicobacter pylori Infection and Incidence of Gastric Cancer Between 1980 and 2022. Gastroenterology, 2024. 166 (4): p. 605-619.
2. Warren, J.R. and B. Marshall, Unidentified curved bacilli on gastric epithelium in active chronic gastritis. Lancet, 1983. 1 (8336): p. 1273-5.
3. Marshall, B.J. and J.R. Warren, Unidentified curved bacilli in the stomach of patients with gastritis and peptic ulceration. Lancet, 1984. 1 (8390): p. 1311-5.
4. Zullo, A., et al., Gastric MALT lymphoma: old and new insights. Ann Gastroenterol, 2014. 27 (1): p. 27-33.
5. Mulder, D.T., et al., Prevalence of Gastric Precursor Lesions in Countries With Differential Gastric Cancer Burden: A Systematic Review and Meta-analysis. Clin Gastroenterol Hepatol, 2024.
6. Malfertheiner, P., et al., Management of Helicobacter pylori infection: the Maastricht VI/Florence consensus report. Gut, 2022.

Introduction

Endoscopic ultrasound (EUS) has broadened its diagnostic and therapeutic scopes, as it is able to target lesions <3mm. High-frequency radial EUS-miniprobes exhibit procedural time reduction, improving image quality and maneuverability. EndoSound Vision System (EndoSound, Oregon, USA) is a new EUS (N-EUS) portable device that fastens onto a standard gastroscope, enabling a free working channel. It is an affordable alternative to conventional-EUS devices; and a promising opportunity to shorten the difficult EUS learning curve.

Aims & Methods

We aim to determine the feasibility and effectiveness of N-EUS in terms of anatomical structures appropriate identification in comparison with conventional-EUS. This prospective study included patients ≥18 years-old undergoing EUS evaluation in June-November 2023 by eight endoscopists from different institutions from South and North America. Patients underwent two consecutive procedures: First, a conventional EUS with a linear-array echoendoscope (Pentax EG38-J1OUT, Pentax Medical, Germany). Then, a different endoscopist performed EUS using the N-EUS coupled to a video gastroscope with therapeutic access (EG29-i10, Pentax Medical, USA). A post-procedure ad-hoc survey was completed by the second endoscopist, to assess N-EUS acceptability in terms of endoscopists’ experience using this device.

Results

Thirty-one patients underwent conventional EUS and N-EUS: 16/31 performed by first-time N-EUS users, 15/31 by endoscopists with prior N-EUS exposure, with median procedural times of 30 and 15 minutes, respectively (p=.04). Main EUS indication was chronic dyspepsia (35.5%). Identified structures during N-EUS were the liver/gallbladder/common bile duct (30/31, 96.8%), full pancreas (29/31, 93.5%), kidneys/left kidney (24/31, 77%; 13/15 endoscopists with prior exposure vs 11/16 first-time N-EUS users; p<.001), left adrenal gland (20/31, 64.5%). Gallbladder (GB) was unobserved in 8 cases, 7/8 had prior cholecystectomy. Evaluation position was 54.8% supine; 3/31 switched to left lateral. All procedures were completed with 100% clinical and technical success, requiring navigation changes in 51.6%, mainly in first-time N-EUS users (p=.001). FNB tissue sampling was achieved in 100% cases. Endoscopists’ experience was predominantly described as “meeting clinical expectations”. The ad-hoc survey showed high psychometrics, with homogeneity among surveyed endoscopists, without redundancy (Cronbach’s alpha of .91), and a good construct validity (Kaiser-Meyer-Olkin Criterion of .728 and Barlett´s test p-value =.0005). When comparing both procedures, appropriate identification of GB or lack of, full pancreas, left/both kidneys and left adrenal gland was 31/31 vs 30/31, 31/31 vs 29/31, 31/31 vs 24/31 (p=.0233), and 29/31 vs 20/31 (p=.0158), respectively. Considering conventional EUS as a gold standard, the observed agreement of N-EUS to identify these structures were 96.8%, 93.5%, 77.4% and 64.5%, correspondingly. Observed agreement for kidneys and left adrenal gland increased when sub-analyzing the cases assessed by endoscopists with prior N-EUS exposure: 81.2% and 76.9%, respectively.

Conclusion

N-EUS is a safe, effective, and feasible tool for EUS evaluation, with high acceptability by endoscopists. Prior ex-vivo training may be warranted for first-time N-EUS users.

References

1. LeBlanc JK. An overview of endoscopic ultrasound equipment. Minerva Gastroenterol Dietol. 2008 Jun;54(2):177-87. PMID: 18319690 .
2. Varas Lorenzo, M.J., Abad Belando, R. and Sánchez-Vizcaíno Mengual, E. (2018), Miniprobe Endoscopic Sonography for Gastrointestinal Tract Assessment: A Case Series of 1451 Procedures. J Ultrasound Med, 37: 293-303. https://doi.org/10.1002/jum.14330
3. Deprez PH. Choice of endosonographic equipment and normal endosonographic anatomy. Best Pract Res Clin Gastroenterol. 2009;23(5):623-37. doi: 10.1016/j.bpg.2009.07.001 . PMID: 19744628 .

Disclosure

Carlos Robles-Medranda is a key opinion leader and consultant for Pentax Medical, Boston Scientific, Steris, Medtronic, Motus, Micro-tech, G-Tech Medical Supply, CREO Medical, EndoSound, and mdconsgroup. All other authors declare no conflicts of interest.

Introduction

Computer-Aided Diagnosis (CADx) systems have become available to assist endoscopists with optical polyp diagnosis. However, current CADx systems have not been trained to recognize sessile serrated lesions (SSLs) and currently classify SSLs incorrectly either as hyperplastic or adenomatous. We aimed to assess the effect of current CADx systems on diagnostic accuracy of SSL recognition among a diverse group of endoscopists.

Aims & Methods

We conducted a prospective video-based study including 17 endoscopists (gastroenterologists and surgeons, experts, and trainees). Each endoscopist reviewed the same video and image series of 115 polyps (33% SL, 33% Hyperplastic Polyps (HP), 34% tubular adenoma (TA)). Histopathological diagnosis of the polyps was used as the gold standard for comparison. Presentation of polyps (SSL, TA, HP) was randomized so that 50% of the polyps seen were presented with displaying a CADx diagnosis for the polyp and 50% were presented without displaying a CADx diagnosis.

Results

A total of 1955 optical polyp diagnoses were performed. The optical diagnostic accuracy amongst various endoscopists to recognize SSLs was: expert gastroenterologists (52.3%, 95% Confidence Interval (CI) 46.3-58.3), trainee gastroenterologists (38.2%, CI 27.2-49.1), expert surgeons (22.1%, CI 16.2-28.0) and trainee surgeons (14.9%, CI 8.4-21.5). Further details are provided in Table 1. CADx identified 52.6% of SSL as HP and 47.4% as TA. CADx identified 60.5% of HP correctly and 82.1% of TA correctly. The optical diagnostic accuracy for HP and TA improved among endoscopists when a correct CADx diagnosis was displayed to the endoscopists compared to when no CADx information was displayed (HP: 63.7% vs 38.4% / TA: 81.5% vs 70.5%; p=0.05). The optical diagnostic accuracy of the endoscopists worsened when the displayed CADx diagnosis was incorrect (HP: 24% / TA: 25%). The diagnostic accuracy of the endoscopists for SSL did not differ significantly when the CADx diagnosis, which is currently untrained to diagnose SSL and can only classify polyps as hyperplastic or adenomatous, was displayed (37.2% vs 33.1%, p=0.05).

	Sessile Serrated Lesion	Hyperplastic Polyp	Adenoma
CADx Alone	HP: 52.6 (36.8-68.5) TA: 47.4 (31.5-63.2)	Sn 60.5 (49.5-71.5) Sp 82.1 (70.0-94.1) PPV 86.8 (77.7-95.9) NPV 51.6 (39.2-64.1)	Sn 82.1 (70.0-94.1) Sp 60.5 (49.5-71.5) PPV 51.6 (39.2-64.1) NPV 86.8 (77.7-95.9)
CADx diagnosis NOT displayed All Endoscopists (n=17)	Sn 37.2 (31.9-42.4) Sp 78.6 (75.5-81.8) PPV 46.2 (40.1-52.2) NPV 71.7 (68.4-75.0)	Sn 38.4 (33.1-43.7) Sp 79.8 (76.8-82.9) PPV 48.4 (42.3-54.6) NPV 72.4 (69.2-75.7)	Sn 70.5 (65.6-75.4) Sp 64.7 (61.0-68.4) PPV 50.6 (46.1-55.2) NPV 81.0 (77.6-84.4)
CADx Displayed Correct Diagnosis All Endoscopists (n=17)	Sn 33.1 (28.0-38.3) Sp 79.4 (76.3-82.5) PPV 44.2 (38.0-50.5) NPV 70.6 (67.3-73.9)	Sn 63.7 (57.4-70.0) Sp 78.7 (74.9-82.5) PPV 60.2 (53.9-66.4) NPV 81.1 (77.4-84.8)	Sn 81.5 (76.9-86.2) Sp 86.5 (83.2-89.9) PPV 80.7 (76.0-85.3) NPV 87.2 (83.9-90.5)
CADx Displayed Incorrect Diagnosis All Endoscopists (n=17)		Sn 24.0 (15.6-32.4) Sp 77.4 (71.7-83.0) PPV 33.3 (22.4-44.2) NPV 68.3 (62.4-74.2)	Sn 25.0 (14.0-36.0) Sp 45.2 (39.1-51.4) PPV 9.8 (5.1-14.5) NPV 71.7 (64.7-78.7)
Breakdown of the performance by endoscopists category GE: Gastroenterologist Expert (n=7) GT: Gastroenterologist Trainee (n=2) SE: Surgery Expert (n=5) ST: Surgery Trainee (n=3)
Overall	GE: 52.3 (46.3-58.3) GT: 38.2 (27.2-49.1) SE: 22.1 (16.2-28.0) ST: 14.9 (8.4-21.5)	GE: 35.0 (29.2-40.7) GT: 61.8 (50.9-72.8) SE: 53.7 (46.6-60.8) ST: 42.1 (33.0-51.2)	GE: 74.7 (69.6-79.9) GT: 64.1 (53.5-74.7) SE: 66.2 (59.5-72.8) ST: 74.4 (66.4-82.3)
CADx diagnosis NOT displayed	GE: 54.1 (45.7-62.6) GT: 39.5 (23.9-55.0) SE: 23.2 (14.7-31.6) ST: 19.3 (9.1-29.5)	GE: 27.8 (20.2-35.4) GT: 57.9 (42.2-73.6) SE: 53.7 (43.7-63.7) ST: 24.6 (13.4-35.7)	GE: 72.8 (65.3-80.3) GT: 66.7 (51.9-81.5) SE: 70.4 (61.4-79.4) ST: 69.5 (57.7-81.2)
CADx Displayed Correct Diagnosis	GE: 50.4 (41.9-58.9) GT: 36.8 (21.5-52.2) SE: 21.1 (12.9-29.3) ST: 10.5 (2.6-18.5)	GE: 50.6 (40.2-60.9) GT: 84.6 (70.7-98.5) SE: 67.2 (55.9-78.4) ST: 73.2 (59.6-86.7)	GE: 85.0 (78.4-91.5) GT: 75.0 (60.0-90.0) SE: 73.4 (63.7-83.2) ST: 91.5 (83.5-99.5)
CADx Displayed Incorrect Diagnosis		GE: 25.0 (12.2-37.8) GT: 25.0 (0.5-49.5) SE: 21.4 (6.2-36.6) ST: 25.0 (3.8-46.2)	GE: 37.5 (18.1-56.9) GT: 0.0 (0.0-0.0) SE: 16.7 (0.0-33.9) ST: 27.3 (1.0-53.6)

Table 1. CADx and endoscopists performance in optical diagnosis of SSL, HP and adenomas depending on the polyp histology in percentages of correct optical diagnoses with 95% confidence intervals. For SSL, CADx is not able to provide correct diagnosis; thus is presented as CADx shown as either HP or TA)
Sn: sensitivity; Sp: specificity; PPV: positive predictive value; NPV: negative predictive value

Conclusion

Optical polyp diagnosis that includes SSL, TA, and HP recognition remains challenging. Current CADx systems when displaying correct diagnostic outputs improve diagnostic accuracy of the endoscopists. CADx systems not trained for SSL recognition do not significantly affect SSL recognition among endoscopists even if an incorrect diagnostic output (TA or HP) is displayed. Development of improved CADx systems with SSL recognition and higher diagnostic accuracy is warranted.

Disclosure

Dr Jeffrey Marks: Consultant for Boston Scientific and Steris Endoscopy

Introduction

Recent investigations have highlighted specific dimensions within the bony pelvis as potential indicators of total mesorectal excision (TME) outcomes. However, manual pelvimetry is labor-intensive.

Aims & Methods

This study endeavors to develop an automated pelvimetry system employing artificial intelligence (AI) to facilitate the preoperative identification of high-risk patients. Specifically, our research aims to construct and validate an AI-based automated pelvimetry tool and assess the prognostic utility of pelvimetric measurements in predicting postoperative complications necessitating reintervention (Clavien Dindo classification ≥ 3) in patients undergoing minimally invasive TME. This study is a multicenter retrospective cohort investigation incorporating clinical and imaging data from 1,700 patients treated at eight experienced TME centers. Leveraging a Deep-Learning network, our algorithm locates multiple anatomical landmarks to quantify pelvic inlet, pelvic outlet, and depth distances accurately. The accuracy of the developed algorithm is compared against inter-observer variability observed in manually conducted pelvimetry. The prognostic significance of pelvimetry for (post-)operative outcomes is evaluated through multivariate regression analyses.

Results

The AI-driven landmark localization achieved a mean margin of error of 5 millimeters across various pelvic dimensions, while manual pelvimetry demonstrated an inter-observer variation of 2.5 millimeters. Analyses of pelvimetric measurements and postoperative complications identified pelvic depth as an independent predictor of such complications (p < 0.001).

Conclusion

An artificial intelligence algorithm could successfully be used for accurate automated pelvic dimension determination. These pelvimetric measurements hold promise as a valuable decision support tool for surgeons in preoperatively assessing the risk of postoperative complications following minimally invasive TME.

Introduction

Potential celiac disease (PCD) is a clinical condition characterized by the presence of a positive CD-specific serology and a normal intestinal architecture. Asymptomatic PCD patients are generally addressed to continue on a gluten containing diet (GCD), but long-term risks of this approach have never been explored.

Aims & Methods

In the present study we aim at investigating nutritional and autoimmune complications possibly developing overtime in a cohort of asymptomatic PCD children left on a GCD. We compared parameters of growth, nutritional status, and autoimmunity between time of diagnosis and in occasion of last medical check, after a long-term gluten containing diet.

Results

Altogether we collected data from 171 PCD children with a mean follow-up time of 3 years (range 0.35-15.3 years). During follow-up, despite patients did not reduce the amount of daily gluten intake , anti-tissue transglutaminase (anti-TG2) antibodies spontaneously and significantly decreased. Most of parameters analysed did not change over follow-up (height centile, ferritin, albumin., cholesterol, calcium, alkaline phosphatase , parathormone, vitamin D, p>0.05) or even improved (weight and BMI centile, haemoglobin, blood iron, HDL, glycaemia and HbA1C, p<0.05), always remaining in the limit of normality. Equally autoantibodies for other concomitant autoimmune disorders did not increase overtime. Similar results were obtained excluding from the analysis patients that stopped producing anti-TG2 and those with a follow up < 3 years.

Conclusion

Our pilot study provides reassuring results on the maintenance of a gluten containing diet in asymptomatic PCD children, even when a long term follow-up is considered.

Disclosure

no conflict of interest to declare

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a lethal disease with the lowest 5-year survival rate of all cancers due to late diagnosis and lack of effective treatment. One of the challenges of developing molecular diagnosis and molecular targeted therapy for PDAC is the “mismatch” between research models and patients.

Aims & Methods

This study aimed i) to identify common proteins among different research models and patients and accordingly ii) to identify potential hub proteins and signaling pathways involved in the PDAC pathogenesis by using proteomics-based system modeling of multimodal datasets derived from multiple research models and patients. Mouse PDAC cells (derived from a genetically engineered mouse model with KrasG12D/Pdx1-Cre), PDAC orthotopically tumors in C57B/L6 mice (n=5), mouse PDAC spheroids, mouse pancreatic organoids, mouse PDAC organoids, normal mouse pancreatic tissues (n=5), human PDAC tumor tissue samples from treatment-naїve (n=20), neoadjuvant chemotherapy (n=22), and human PDAC organoids were included. Mass spectrometry-based proteomics, Rstudio, Cytoscape 3.10 and pathway enrichment analysis (using ingenuity pathways analysis, including human PDAC transcriptomic in addition) were used.

Results

In the comparison of PDAC models and patients, shared proteins were identified as follows: 60% between the mouse PDAC cells and the mouse PDAC spheroids, 78% between the mouse PDAC cells and the mouse PDAC tissue, 56% between the mouse PDAC spheroids and the mouse PDAC tissue, 66% between the mouse PDAC cells and the mouse PDAC organoids, 54% between the mouse PDAC spheroids and the mouse PDAC organoids, 38% between the mouse PDAC and the human PDAC tissues, 31% between the human PDAC tumor and the human PDAC organoids, and 60% between the mouse and the human PDAC organoids. A total of 1,367 common proteins were identified within all the PDAC models. Among these proteins ACTB, HSP90AA1, HSPA8, HSP90AB1, HSPA4, EEF2, RPS3, RPS20 and GAPDH were the top 10 hub proteins (with >300 node connections). Upon comparison of all eight models, eight proteins were identified as unique to PDAC, namely FAHD1, ARSA, NEK9, SERPINB9, LAMTOR2, SNRPG, TMED3, and RAB1A. There was a correlation in terms of z-score regarding signaling pathways between the mouse PDAC and human PDAC (p<0.001), particularly in activations of nerve growth factor signaling, integrin signaling, and neuregulin signaling. Central hub proteins commonly found in the mouse models included vinculin (membrane-cytoskeletal protein) as one of the high connectivity nodes and β-catenin and probable ubiquitin carboxyl-terminal hydrolase FAF-X (involved in axonal growth and neuronal cell migration) as high centrality nodes. In patients, central hub proteins that were dysregulated after chemotherapy included 14-3-3 protein zeta/delta (encoded by YWHAZ gene), β-catenin and SRC. Among the 47 proteins that were associated with altered metabolism after the chemotherapy in patients, CD44, CTNND1, APOA 1, PARVA, and SCRBS1 were nodes with low degrees, and desmosomes (DSP, KP3, PPL), PRTN3, thrombospondin-2 (THBS2), and DMBT1 were the most down-regulated proteins without nodes, and APOA1, caveolin-1 (encoded by CAV1), LRP1 were the most up-regulated proteins with nodes.

Conclusion

The system modeling identified the common proteins among the research models and patients and the hub proteins and signaling pathways, suggesting the potential translational targets of PDAC.