Miss Rate of Gastric Neoplasms Under Computer-aided Endoscopy

Computer-aided Gastric Lesion Localization and Miss Rate of Gastric Neoplasms: a Tandem, Randomized Controlled Study

This prospective randomized trial compares AI-assisted upper gastrointestinal endoscopy with high definition upper gastrointestinal endoscopy in term of missed rate of gastric neoplasm. The investigators hypothesize the miss rate of high definition upper gastrointestinal endoscopy is higher than AI-assisted upper gastrointestinal endoscopy.

Study Overview

• Status: Recruiting
• Conditions: Artificial Intelligence; Gastric Neoplasm
• Intervention / Treatment: Device: AI-assisted upper gastrointestinal endoscopy

Detailed Description

Patients will be randomly assigned to begin with AI-assisted upper gastrointestinal endoscopy follow immediately by high definition (HD) upper gastrointestinal endoscopy (AI-HD group); or start with HD upper gastrointestinal endoscopy follow immediately by AI-assisted upper gastrointestinal endoscopy (HD-AI group). The random allocation sequence is generated by a computer-generated random numerical series, with 1 representing the AI-HD group and 0 representing the HD-AI group. Randomization is conducted in blocks of four at a 1:1 ratio stratified by indications (screening/surveillance vs others), endoscopist's experience (experienced versus less experienced) and mode of sedation (unsedated vs sedated). Experienced endoscopist is defined as qualified endoscopists with more than 7 years experience in upper endoscopy, whereas less experienced endoscopists include fellows and trainees. A research assistant, not directly involved in this study, maintained all randomization codes which are contained within individual opaque envelopes. Upon obtaining patient consent, the envelope will be opened to reveal the assigned examination sequence. Patients remain blinded to their group allocation throughout the study, but the performing endoscopist is aware of the assigned allocation.

Participating endoscopists will receive training on the interpretation of real-time AI detection system as well as detection of dysplasia under HD endoscopy before performing study. All patients will fast for at least 6 hours before the procedure. All examinations will be performed with HD endoscopes (ELUXEO 7000 video system, Fujifilm Co, Tokyo, Japan) under white light. The artificial intelligence assisted gastric dysplasia localization system uses a graphical user interface for real-time display of lesion detection with bounding boxes (Fujifilm Co, Tokyo, Japan).

Each eligible patient will undergo a same-day tandem upper gastrointestinal endoscopy performed by the same endoscopist to evaluate the miss rate of gastric neoplasm. Patients first receive either AI-assisted or HD upper gastrointestinal endoscopy under white light endoscopy, immediately followed by cross-over to other procedure. Endoscopists will be assisted by a research assistant (RS), who activates or deactivates the lesion detection function of AI system between the two examinations. Both first and second examinations are conducted in accordance with the systematic gastric screening protocol, and only the gastric cavity was rescanned during the second observation. The minimal inspection time of the stomach should be 3 minute for the both examination.

Biopsies of all targeted lesions will be taken at the end of each examination. Endoscopists are instructed to biopsy lesions meeting the following criteria in HD examinations: color differences, loss of vascularity, slight elevation or depression, nodularity, thickening, abnormal convergence or flattening of folds, irregular margins, irregular discoloration, or irregular surface. During AI-assisted examinations, targeted lesions are defined as focal lesions marked in localization boxes. Endoscopists are instructed to biopsy areas stably marked with localization boxes that persisted for 5 seconds by the AI system.

• Study Type: Interventional
• Enrollment (Estimated): 1000
• Phase: Not Applicable

Contacts and Locations

• Study Contact: Name: Thomas Ka-Luen Lui; Phone Number: +852 97360997; Email: tkllui@hku.hk

Study Locations

• Hong Kong
  • Hong Kong, Hong Kong
    • Recruiting
    • Queen Mary Hospital, The University of Hong Kong
    • Contact: Thomas Ka Luen Lui; Phone Number: +852 97360997; Email: tkllui@hku.hk

Participation Criteria

Eligibility Criteria

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

Description

Inclusion Criteria:

• Patients aged 40 or older
• Scheduled for elective upper endoscopy

Exclusion Criteria:

• Pregnant women,
• Inability to provide written informed consent
• Prior gastrectomy, and
• Patients deemed unsuitable or high-risk for endoscopy with severe comorbid illnesses

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Treatment
• Allocation: Randomized
• Interventional Model: Crossover Assignment
• Masking: None (Open Label)

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: AI-HD group; AI-assisted upper gastrointestinal endoscopy follow immediately by high definition (HD) upper gastrointestinal endoscopy	Device: AI-assisted upper gastrointestinal endoscopy; AI-assisted upper gastrointestinal endoscopy
Active Comparator: HD-AI group; HD upper gastrointestinal endoscopy follow immediately by AI-assisted upper gastrointestinal endoscopy	Device: AI-assisted upper gastrointestinal endoscopy; AI-assisted upper gastrointestinal endoscopy

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Gastric neoplasia miss rate	the number of newly detected gastric neoplasia in the second examination divided by the total number of gastric neoplasia detected in both examinations for each patient.	during the intervention

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Gastric neoplasia detection rate	The proportion of patients with one or more gastric neoplasms detected in the first examination	during the intervention
The biopsy rate	The proportion of patients who had a biopsy sample taken as a result of the first examination	during the intervention
The number of gastric neoplasms per patient	The total number of gastric neoplasms detected in the first examination divided by the total number of patients	during the intervention
The miss rate of patients with gastric neoplasms	The number of patients with a newly detected gastric neoplasm in the second examination divided by the total number of patients with gastric neoplasms detected in both examinations	during the intervention
The positive predictive value (PPV) for gastric neoplasms	The number of detected gastric neoplasms divided by the number of all targeted lesions in the first examination	during the intervention
Inspection time	The procedure time of the examinations minus the time taken to obtain a biopsy sample, measured from the passage of the endoscope through the gastroesophageal junction until the completion of the whole examination of the gastric cavity, excluding the inspection time of the duodenum and the biopsy time	during the intervention
The macroscopic and histological types of detected and missed neoplasia in the first examination	The macroscopic and histological types of detected and missed neoplasia in the first examination	during the intervention
Number of biopsy per patient in the first arm	Number of biopsy per patient in the first arm	during the intervention
Percentage of positive biopsy	The number of gastric neoplasm detected divided by total number of biopsy in the first arm	during the intervention

Collaborators and Investigators

• Sponsor: The University of Hong Kong
• Investigators: Principal Investigator: Thomas Ka-Luen Lui, The University of Hong Kong

Study record dates

Study Major Dates

• Study Start (Actual): November 1, 2024
• Primary Completion (Estimated): September 1, 2026
• Study Completion (Estimated): December 31, 2026

Study Registration Dates

• First Submitted: June 25, 2024
• First Submitted That Met QC Criteria: July 7, 2024
• First Posted (Actual): July 11, 2024

Study Record Updates

• Last Update Posted (Actual): November 20, 2025
• Last Update Submitted That Met QC Criteria: November 19, 2025
• Last Verified: November 1, 2025

More Information

Terms related to this study

• Keywords: Artificial intelligence; Gastric Neoplasm; Miss rate
• Additional Relevant MeSH Terms: Neoplasms by Site; Neoplasms; Gastrointestinal Neoplasms; Digestive System Neoplasms; Digestive System Diseases; Gastrointestinal Diseases; Stomach Diseases; Stomach Neoplasms
• Other Study ID Numbers: protocol_upper_RCTV10

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

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