Management of Pancreatic Cystic Lesions Using Artificial Intelligence Based on EUS and Multimodal Data

A Multimodal Artificial Intelligence Model for Subtyping Diagnosis and Clinical Management of Pancreatic Cystic Lesions Based on Endoscopic Ultrasound and Clinical Information

The primary objective is to construct a multimodal AI model (Cyst-AI) based on EUS images and clinical data such as imaging features(CT or MRI) and laboratory tests to assist endoscopists in the diagnosis of pancreatic cystic lesions(PCLs), mainly differentiating mucinous from non-mucinous lesions.

The secondary objective is to evaluate the model's effectiveness in risk stratification and clinical management for patients with PCLs.

Study Overview

• Status: Recruiting
• Conditions: Mucinous Cystadenoma of Pancreas; Pseudocyst Pancreas; Serous Cystadenoma; Neuroendocrine Tumors, NET; Pancreatic Cystic Lesion; Intraductal Papillary Mucinous Neoplasm of Pancreas
• Intervention / Treatment: Diagnostic test: Cyst-AI model

Detailed Description

With the development of medical imaging technology, the detection rate of pancreatic cystic lesions (PCLs) has been increasing notably. Although most cysts are benign, a considerable subset has the potential for malignant transformation. Clinical management is based on diagnosis and risk stratification. For PCLs，different diagnosis and risk stratification lead to entirely different clinical strategies and outcomes, which are closely related to the quality of life, economic burden, and psychological stress of patients. Endoscopic ultrasound (EUS) has played a crucial role in the further differential diagnosis of PCLs. Artificial intelligence (AI) has also shown great potential in clinical diagnosis and management. Thus, we plan to retrospectively collect patients' EUS imaging data, radiological and laboratory tests, and other clinical information to construct a model named Cyst-AI which integrates the function of diagnosis and clinical management, to assist in clinical decision-making.

• Study Type: Observational
• Enrollment (Estimated): 500

Contacts and Locations

• Study Contact: Name: Bin Cheng; Phone Number: 86-13986097542; Email: b.cheng@tjh.tjmu.edu.cn

Study Locations

• China
  • Hubei
    • Wuhan, Hubei, China, 430030
      • Recruiting
      • Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology
      • Contact: Bin Cheng; Phone Number: +8613986097542; Email: b.cheng@tjh.tjmu.edu.cn
    • Wuhan, Hubei, China, 430030
      • Not yet recruiting
      • Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology
      • Contact: Bin Cheng; Phone Number: 86-13986097542; Email: b.cheng@tjh.tjmu.edu.cn

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

• Sampling Method: Non-Probability Sample

Study Population

The cohort will be selected from several hospitals in China, including Tongji Hospital, Tongji Medical College, HUST.

Description

Inclusion criteria:

• Patients whose EUS results indicates pancreatic cystic or cystoid lesions;
• Mucinous lesions: including mucinous cystic neoplasm (MCN), intraductal papillary mucinous neoplasm (IPMN);
• Non-mucinous lesions: including pancreatic pseudocyst, serous cystic neoplasm (SCN), cystic neuroendocrine tumor (cNET).

Exclusion criteria:

• Patients whose age is less than 18 years old;
• Patients who have undergone pancreatic surgery before the EUS examination;
• Patients who have received chemotherapy and radiotherapy for pancreatic tumors before the EUS examination;
• Pathological results indicate that pancreatic lesions are metastatic lesions from other sites;
• Patients whose EUS images or reports are missing;
• EUS image quality does not meet the requirements for review, such as blurry imaging or containing artifacts, biopsy needles, measuring scales, or other additional annotations that are not part of the original EUS image;
• Patients whose final diagnosis is unclear.

Study Plan

How is the study designed?

Number of groups / cohorts

1

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Cyst-EUS; Patients before 2026 with EUS pictures of pancreatic cystic lesions or cystoid-material lesions have been included in this cohort.	Diagnostic test: Cyst-AI model; The multi-center collected data will be divided into a training set, a validation set, and a test set for developing and testing the cyst-AI model.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
The performance of the diagnostic model in differentiating mucinous from non-mucinous PCLs	The performance of the Cyst-AI diagnostic model will be evaluated using the area under the receiver operating characteristic curve (AUC-ROC), with sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) calculated from the model's predictions on the independent validation dataset. PCLs: pancreatic cystic lesions.	Within 3 months upon completion of the diagnostic model training.
The risk stratification performance of the clinical management model for mucinous PCLs	The performance of the Cyst-AI risk stratification model to correctly classify lesions into "low risk", "intermediate risk" and "high risk", will be evaluated using the area under the receiver operating characteristic curve (AUC-ROC), with sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) calculated from the model's predictions on the independent validation dataset. PCLs: pancreatic cystic lesions.	Within 3 months upon completion of the risk stratification model training.

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
The performance of the diagnostic model in differentiating specific types of PCLs	The performance of the Cyst-AI diagnostic model will be evaluated using the area under the receiver operating characteristic curve (AUC-ROC), with sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) calculated from the model's predictions on the independent validation dataset. PCLs: pancreatic cystic lesions.	Within 3 months upon completion of the diagnostic model training.
The clinical management performance of the clinical management model for mucinous PCLs	The performance of the Cyst-AI clinical management model to provide accurate clinical recommendations, will be evaluated using the area under the receiver operating characteristic curve (AUC-ROC), with sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) calculated from the model's predictions on the independent validation dataset. PCLs: pancreatic cystic lesions.	Within 3 months upon completion of the clinical management model training.
The performance of the model in assisting endoscopists of different levels in diagnosing and managing PCLs	The performance of the Cyst-AI model in assisting endoscopists will be evaluated using the area under the receiver operating characteristic curve (AUC-ROC), with sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) calculated from the model's predictions on the independent validation dataset. PCLs: pancreatic cystic lesions.	Within 1 months upon completion of the human-machine confrontational crossover study
The impact of the model on the decision-making process of endoscopists	Questionnaire for endoscopists after assessment will be used to evaluate the degree of impact.	Within 1 months upon completion of the human-machine confrontational crossover study.

Collaborators and Investigators

• Sponsor: Huazhong University of Science and Technology

Study record dates

Study Major Dates

• Study Start (Actual): January 1, 2025
• Primary Completion (Estimated): April 1, 2026
• Study Completion (Estimated): June 1, 2026

Study Registration Dates

• First Submitted: December 9, 2025
• First Submitted That Met QC Criteria: March 5, 2026
• First Posted (Actual): March 11, 2026

Study Record Updates

• Last Update Posted (Actual): March 11, 2026
• Last Update Submitted That Met QC Criteria: March 5, 2026
• Last Verified: March 1, 2026

More Information

Terms related to this study

• Keywords: artificial intelligence; endoscopic ultrasound; risk stratification; multimodal; differentiation; clinical management; pancreatic cystic lesions
• Additional Relevant MeSH Terms: Endocrine System Diseases; Neoplasms by Site; Neoplasms; Neoplasms by Histologic Type; Digestive System Neoplasms; Digestive System Diseases; Endocrine Gland Neoplasms; Pancreatic Diseases; Neoplasms, Glandular and Epithelial; Adenoma; Neuroectodermal Tumors; Neoplasms, Germ Cell and Embryonal; Neoplasms, Nerve Tissue; Cysts; Neoplasms, Cystic, Mucinous, and Serous; Neoplasms, Ductal, Lobular, and Medullary; Pancreatic Neoplasms; Pancreatic Cyst; Cystadenoma; Pancreatic Intraductal Neoplasms; Neuroendocrine Tumors; Pancreatic Pseudocyst; Cystadenoma, Serous
• Other Study ID Numbers: Cyst-AI 2024

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No