Status
Conditions
Treatments
About
The goal of this clinical trial is to evaluate the diagnostic accuracy of the AI-integrated Capsule Gastroscopy (ACG) system in simulated home-use conditions for detecting upper gastrointestinal (UGI) abnormalities. It will also compare the diagnostic accuracy and time efficiency of AI-assisted interpretation versus manual reading of ACG data. The main questions it aims to answer are:
What is the diagnostic accuracy of the ACG system, using conventional esophagogastroduodenoscopy (EGD) as a standard of reference?
Does AI-assisted ACG reading improve diagnostic accuracy or reduce reading time compared to manual ACG video reading?
Researchers will compare ACG results to conventional EGD findings (standard of reference) to determine if ACG can serve as a reliable method for UGI disease detection in home scenarios.
Participants will:
Undergo an ACG examination in a simulated home environment. Complete an EGD procedure within 24 hours post-ACG ingestion.
Full description
Gastroscopy stands as the gold standard for diagnosis of upper gastrointestinal (UGI) diseases (1, 2). However, China faces a significant challenge due to its large population base and an insufficient number of gastrointestinal endoscopists, which limits the widespread adoption of conventional electronic gastroduodenoscopy (EGD). Additionally the discomfort associated with EGD and the low compliance rate reduce its effectiveness in meeting public health needs (3).
Capsule endoscopy (CE), primarily comprising magnetic capsule endoscopy (MCE) and powerless capsule endoscopy (PCE), is an innovative medical technology that enables comfortable and comprehensive examinations of the digestive tract (4-6). MCE relies on costly magnetic guidance equipment for capsule manipulation (7), a process that requires specialized technicians. These inherent limitations pose barriers to its clinical use and acceptance in various settings, such as primary hospitals and community health centers.
To overcome these challenges, we have developed an automated, wireless, artificial intelligent (AI) integrated Capsule Gastroscopy (ACG) System (GICE-1000, AI Mobile Gastroscopy, Guangzhou Side Medical Technology Co., Ltd) for detecting gastric lesions. After the stomach is distended with ingested, the capsule examines the entire stomach through standardised body position changes by the participant, eliminating the need for any magnetic guidance equipment. The video sequence can be viewed in real-time using a cellphone and transmitted via WIFI to a cloud server for remote reading. Characterized by its comfort, operational simplicity, and remote controllability, GICE-1000 is poised to enhance the early detection and treatment rate of UGI abnormalities in different settings, including both community hospitals and homes, thereby alleviating healthcare system burdens. However, there is no prospective study to assess the diagnostic accuracy of GICE-1000 in home scenarios.
The study is structured into three distinct phases: the screening period, the examination period, and the follow-up period.
Screening Period:
The investigators or research assistants will identify eligible participants, including patients and volunteers in the hospital, and provide a detailed explanation of the study, including relevant information and potential risks.
Once a participant signs the informed consent form, the researcher will assign a screening number, record baseline demographic information, and evaluate eligibility based on the study's inclusion and exclusion criteria. Participants who meet the inclusion criteria and do not meet any exclusion criteria will be assigned an enrollment number.
Intervention Period:
2.1 Preparation One day prior to the examination, investigators or research assistants will instruct enrolled participants on preparation for the ACG examination. Participants will be advised to fast for 8 hours before the procedure but may drink clear, non-carbonated beverages.
2.2 Procedure of ACG examination
2.3 EGD examination
2.4 ACG reading The ACG video will first be reviewed by a capsule reader (with experience reading >100 capsules) who is blinded to the EGD results at the center where the patient was enrolled (non-randomized). The initial reading will be conducted in standard mode, according - at 10 frames per s in single-view mode in the small bowel, and 20 frames per s in the oesophagus or stomach. Landmarks, including the first image of the gastrointestinal tract, the first stomach image, the first duodenal image, and representative images of anatomical structures (including esophageal, EGJ, gastric fundus, gastric angle, cardia, body, antrum, pylorus, 1st and 2nd part of the duodenum), were manually selected by the reader. Observed findings were recorded through mouse clicks, ensuring comprehensive documentation of the anatomical and pathological features identified during the video review. This systematic approach allowed for detailed tracking and analysis of abnormalities and structure coverage. The reader should record any abnormalities noted during the ACG video reading, and take at least one representative image of each lesion. The recorded data will include lesion location, morphology, number of lesions, type of lesion, a visual estimation of lesion size, and coverage of gastric anatomical structures (fundus, gastric angle, cardia, body, antrum, pylorus). Additional data will include timestamps captured at the start and end of the reading, image quality, gastric cleanliness, and the degree of gastric filling, among other factors.
The ACG video will also be anonymized and randomly allocated to another center for blinded AI-assisted reading, with a reader (with experience reading >100 capsules) who is unaware of the results from the initial manual reading and the EGD results. The reader will analyze the images and video data provided by the AI platform (Endonet) to make an AI-assisted diagnosis. Following the same documentation protocol as the manual reading, landmarks such as the first images of the gastrointestinal tract, stomach, and duodenum, as well as key anatomical regions (e.g., esophagus, EGJ, fundus, gastric angle, cardia, body, antrum, pylorus, D1 and D2) will be manually identified and captured through mouse clicks if available. Observations, abnormalities, and representative lesion images (at least one image for each lesion) will be recorded, along with lesion characteristics (location, morphology, number of lesions, type of lesion, estimated size), coverage of gastric anatomical structures (fundus, gastric angle, cardia, body, antrum, pylorus), overall image quality, cleanliness, and gastric filling, ensuring thorough documentation and analysis.
The study is aim to compare the diagnostic accuracy and reading time of AI-assisted ACG reading with standard manual reading. The documented data will include lesion location, morphology, number of lesions, type of lesion, size, coverage of gastric anatomical structures (fundus, gastric angle, cardia, body, antrum, pylorus), as well as image quality, gastric cleanliness, and degree of gastric filling. Additionally, the number of AI-selected images and AI-assisted ACG reading time will be recorded. The diagnostic outcomes of EGD will serve as the reference standard for assessing the diagnostic accuracy of ACG (including AI-assisted ACG reading).
Enrollment
Sex
Ages
Volunteers
Inclusion and exclusion criteria
Inclusion criteria
Aged 18 years or older.
Individuals meeting the following criteria:
i. Healthy volunteers; ii. Suspected presence of gastrointestinal diseases, with one or more of the following clinical symptoms: abdominal pain, nausea, vomiting, hematemesis, black or bloody stools, loss of appetite, bloating, or indigestion; iii. Follow-up of gastric lesions post-endoscopic resection.
Willing to participate voluntarily in the clinical trial and provide written informed consent.
Capable of communicating with researchers and complying with trial requirements.
Exclusion criteria
Primary purpose
Allocation
Interventional model
Masking
482 participants in 2 patient groups
Loading...
Central trial contact
Xiaobei Luo
Data sourced from clinicaltrials.gov
Clinical trials
Research sites
Resources
Legal