Introduction

Accurate diagnosis of pancreatic lesions poses a considerable challenge due to the involvement of multiple examinations in the diagnostic process. Endoscopic ultrasound (EUS) stands out as a valuable tool enabling characterization and tissue sampling of both cystic and solid lesions, thereby influencing treatment decisions and follow-up strategies significantly. Pancreatic cystic neoplasms (PCNs) encompass mucinous (M-PCN) and non-mucinous lesions (NM-PCN), while pancreatic adenocarcinoma (P-ADC) remains the most prevalent cause of pancreatic solid lesions (PSL), trailed by pancreatic neuroendocrine tumors (P-NET). Despite the potential benefits of artificial intelligence, particularly deep learning algorithms, in EUS, research on the role of convolutional neural networks (CNNs) in pancreatic lesions is scarce.

Aims & Methods

This multicentric study aims to develop a CNN tailored for the detection and differentiation of PCNs (specifically M-PCN and NM-PCN) as well as PSLs (particularly P-ADC and P-NET). Our group developed a retrospective multicenter study based on patients submitted to EUS for pancreatic lesion characterization. A trinary CNN differentiated normal pancreas tissue from M-PCN and NM-PCN. A binary CNN distinguished P-ADC from P-NET. The total dataset was divided into a training and testing dataset (used for model’s evaluation) in a 90/10% ratio. The model performance was evaluated through its sensitivity, specificity, positive and negative predictive values and accuracy. The Sci-kit learn v0.22.2 was used for statistical analysis of the model.

Results

A total of 378 videos of EUS performed for evaluation of pancreatic lesions during November 2017 and December 2023 in four reference centers (Hospital Universitario Puerta de Hierro Majadahonda, n=234; Centro Hospitalar Universitário São João, n=33; New York University Langone and Manhattan Hospital, n=60; Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo, n=51) were used for development of the model. A total of 126.000 frames were extracted for development of CNN: 64.286 of P-ADC, 29.153 of P-NET, 4.858 of normal pancreas, 19.528 of M-PCN and 8.175 of NM-PCN.
Pancreatic Cystic Neoplasms.
The trinary CNN was developed in order to distinguish between normal pancreas images and cystic lesions, differentiating M-PCN from NM-PCN. The deep learning model identified normal pancreas images with 98.6% sensitivity, 99.2% specificity and 99.1% accuracy. M-PCN were detected with 98.9% sensitivity, 99.1% specificity, a PPV of 99.4% and NPV of 98.3%, with a global accuracy of 99%. Finally, the testing dataset of our CNN revealed 99.3% sensitivity, 99.9% specificity, 99.9% PPV, 99.8% NPV and 99.8% accuracy for the diagnosis of NM-PCN.
Pancreatic Solid Lesions
The binary CNN was developed to differentiate between the two commoner PSL, namely P-ADC and P-NET. The testing dataset revealed 94.0% accuracy for the distinction between the lesions, with 98.7% and 83.7% sensitivities for P-ADC and PNET, respectively.

Conclusion

Our group developed a deep learning model capable of differentiating the main types of pancreatic solid lesions, as well as pancreatic cystic lesions, with high performance values. To our knowledge this is the first multicentric approach for the development of an AI algorithm designed to simultaneously characterize cystic and solid pancreatic lesions.. The development of deep learning models may help differentiate pancreatic lesions.Larger multicentric and prospective studies are needed for the validation and future clinical implementation of this technology.

References

Kuwahara T, Hara K, Mizuno N, et al. Current status of artificial intelligence analysis for the treatment of pancreaticobiliary diseases using endoscopic ultrasonography and endoscopic retrograde cholangiopancreatography. DEN Open. 2024;4(1):e267.

Introduction

The Percutaneous Endoscopic Gastrostomy (PEG) tube is used to provide enteral nutrition when oral intake is not adequate.
Regular follow-up of PEG patients is essential to detect complications and clarify doubts. A partnership protocol was developed with ACES Lisboa Central (ACESLC) and Loures/Odivelas (ACESLO) to improve continuity of care.

Results & Findings

We analyzed data from PEG patients followed between 2023 and 2024. Inclusion criteria for referral to ACES included:
• Clinical stabilization;
• Absence of PEG-related complications requiring hospital admission;
• Willingness and ability of the patient/caregiver to participate actively in care.
Monitoring of the protocol was based on quarterly evaluation of five performance indicators:
1st: Percentage of patients referred to ACES, out of the total seen in PEG review consultations;
2nd: Percentage of referred patients who returned to hospital before the expected date due to complications;
3rd: Percentage of cases in which complete referral documentation, as per protocol, was sent to ACES;
4th: Percentage of cases with confirmation that ACES sent back all required clinical data;
5th: Percentage of compliance with scheduled PEG follow-up appointments at ACESLO.

Summary & Discussion

The 1st indicator shows low referral rates in both years, mainly due to out-of-area institutionalization, clinical worsening, or death.
The 2nd indicator indicates fewer unplanned hospital readmissions among referred patients; when they occurred, complications were the cause.
Results show fewer complications and emergency visits, reflecting the benefit of close ACES collaboration.
The 3rd indicator reflects 100% compliance in sending referral documentation.
The 4th indicator showed gaps in home visit records in 2023, improved in 2024 through better coordination.
One challenge was the mismatch between designated ACES nurses and those conducting home visits, which required articulation with family health teams (USF).
The 5th indicator confirms overall adherence to scheduled PEG reviews.

Conclusion

The protocol streamlined follow-up for PEG patients.
In 2024, a previous limitation was overcome, allowing referrals of institutionalized patients within the protocol’s coverage.
As a pioneering project, we aim to share these findings to support protocol expansion to other units.

References

Kurion, M., Andrews, R., Tattersall, R., McAlindon, M., Wong, E., Johnston, A., Hoeroldt, B., Dear, K. & Sanders, D. (2017). Gastrostomies preserve but do not increase quality of life for patients and caregivers. Clinical Gastroenterology and Hepatology. 15, 1047-1054.
Lee, C.,Kang, H., Lim, Y., Lee, J., Koh, M., Lee, J., Yang, C. & Kim, J. (2013). Comparison of complications between endoscopic and percutaneous replacement of percutaneous endoscopic gastrostomy tubes. JKoren Medical Sciences. 28, 1781-1787.
Lucendo, A. J. & Friginal-Ruiz, A. B. (2014). Percutaneous endoscopic gastrostomy: An update on its indications, management, complications, and care. Revista Española de Enfermedades Digestivas. 106(8), 529-539.

Introduction

EUS-guided Gastroenterostomy (EUS-GE) represents a recent breakthrough in therapeutic endosonography, with most existing data derived from retrospective studies focused on short-term technical outcomes. Limited information is available on long-term outcomes, the need for reinterventions and modifications of Lumen Apposing Metal Stents (LAMS) over time.

Aims & Methods

All consecutive EUS-GE performed in a single institution with the Wireless Simplified Technique (WEST) between 2020 and 2024 were enrolled in a prospective registry (PROTECT, ClinicalTrials.gov NCT04813055), with efficacy and safety outcomes assessed every 60 days. The primary outcomes were symptom recurrence after initial clinical success (rate, proportion) and dysfunction-free survival (DyFS) estimated using Kaplan Meier curves. Endoscopic LAMS reassessments were used to calculate the time to first modification of the stent and surrounding tissues, whilst Radiologic imaging was used to estimate the change in stent calibre (linear regression).

Results

164 EUS-GEs (median age 66 [60-74], male 52.4%, malignant indication 99.4% [pancreatic cancer 73.2%, metastatic 58.3%]) were included. A 20 mm LAMS was used in 98.8%, dilated up to 18 mm in 98.8% of cases. Technical and Clinical success rates were 98.8% and 96.3%, respectively. Adverse events (AEs) were registered in 15.2% (procedure-related 6.1%, severe/fatal 3.7%).
Over a median follow-up of 112 [40-190] days, symptom recurrence occurred in 9.1%, but was related to stent dysfunctions in 10/154 (6.5%) cases. Stent dysfunctions happened after a median of 231 [67-321] days from the procedure, all of which (100%) were successfully managed endoscopically. Estimated DyFS at 6 and 12 months was 97.5% and 82.6%, respectively.
Among 38 patients undergoing endoscopic LAMS re-evaluation for different reasons, tissue reactions on the gastric or jejunal side were observed in 34.2% after a median of 190 [33-351] days, and LAMS ingrowth was seen in 13.2% after a median of 421 [250-560] days. A total of 132 CT scans from 75 patients were analysed, with regression analysis revealing a reduction in LAMS diameter of ≈ 0.25 mm + 0.03 every 2 months (F-test p<.0001).

Conclusion

In the natural history of EUS-GE, subclinical LAMS changes and tissue reactions can be frequently detected during follow-up along with a progressive narrowing of LAMS diameter. However, most of these modifications are asymptomatic, with a 6.5% rate of endoscopic reinterventions, all performed successfully. Analysis of these curves may be helpful to plan scheduled reinterventions, especially in cases with benign indication or in long-term survivors.