An Artificial Intelligence System for Multimodal, Multi-class Diagnosis of Pancreatic Cystic Lesions Based on Endoscopic Ultrasonography

An Artificial Intelligence System for Multimodal, Multi-class Diagnosis of Pancreatic Cystic Lesions Based on Endoscopic Ultrasonography

The aim of this study is to develop and validate an artificial intelligence system named iEUS-PCL (intelligent endoscopic ultrasound system-pancreatic cystic lesions） for detecting and multimodal, multi-class diagnosing pancreatic cystic lesions (PCL) during endoscopic ultrasound (EUS) examination.

This multicenter, prospective cohort study aims to develop and validate a multimodal artificial intelligence system named iEUS-PCL for the detection and differential diagnosis of PCL. The model was developed based on retrospectively collected EUS images, EUS features, clinical data and radiological imaging features of patients who underwent EUS examination. The diagnostic performance of iEUS-PCL will be evaluated prospectively in real-time EUS videos and compared with endosonographers' performance.

Inclusion Criteria:

1. Patients aged ≥18 years scheduled for EUS with suspected pancreatic cystic lesions based on clinical symptoms, medical history, laboratory tests or radiological examinations, and who agree to participate in the research and voluntarily sign the informed consent.
2. Patients with no prior history of treatment for pancreatic lesions.

Exclusion Criteria:

1. Patients with absolute contraindications to EUS examination.
2. Pregnancy or lactating.
3. Uncorrectable coagulopathy (PTT>50 seconds or INR>1.5) and/or uncorrectable thrombocytopenia(platelet count<50×109/L).
4. Upper gastrointestinal obstruction.
5. Patients who underwent surgical treatment or anatomical alterations of the pancreas due to lesions in other thoracic and/or abdominal organs, as well as patients with congenital anatomical abnormalities.
6. Patients who have undergone biliary/pancreatic duct stent placement.
7. Patients who refuse to sign the informed consent.