A Prospective Study to Evaluate the Diagnostic Accuracy of Computer-aided Diagnosis (CADx) System in Real-time Characterization of Colorectal Neoplasia

The Chinese University of Hong Kong

Status: Not yet enrolling

Conditions: Colorectal Neoplasms

Treatments: Device: CADx

Study type: Observational

Funder types: Other

Identifiers

NCT05414383

2022.160

Details and patient eligibility

About

The investigators hypothesize that a newly developed CADx system will have a higher diagnostic accuracy in predicting histopathology of colorectal neoplasia than both expert and junior endoscopists.

Full description

Accurate diagnosis and characterization of colorectal polyps is essential before endoscopic resection. Optical diagnosis by enhanced imaging modality (e.g. Narrow Band Imaging, NBI) allows real-time prediction of histopathology. It can assist endoscopists to select the appropriate technique and differentiate between neoplastic or non-neoplastic polyps. Nevertheless, due to the substantial inter-observer variability, the widespread use was limited.

Recently, artificial intelligence and computer-aided polyp diagnosis (CADx) systems have evolved rapidly. The major limitation was the heterogeneity from different types of imaging modalities. Endocytoscopic images require extra steps for pre-staining and magnification, which are time consuming and operator dependent. As a result, it limits the generalisability and applicability in real-world settings.

A novel CADx system will be developed for real-time histopathological prediction of colorectal neoplasia, by using non-magnified conventional white-light and image enhanced endoscopy (NBI). The diagnostic accuracy of this CADx system will be compared with both expert and junior endoscopists.

Enrollment: 510 estimated patients

Sex: All

Ages: 18+ years old

Volunteers: No Healthy Volunteers

Inclusion criteria

1. They have received colonoscopy for screening, surveillance or symptom investigation;
2. They have endoscopic images and videos captured and stored during colonoscopy which are available to be retrieved;
3. They have histologically proven colorectal neoplasia.
4. Written consent obtained

Exclusion criteria

1. Poor quality endoscopic images and videos defined as:

   1. Incomplete visualization of the colorectal neoplasia due to technical reasons (e.g. out-of-focus, motion-blurred or insufficient illumination);
   2. Artifacts due to mucus, air bubbles, stool, or blood.

2. Active gastrointestinal bleeding;

3. Fulminant colitis;

4. Obscured view due to poor bowel preparation;

5. Artificial staining of lesion due to chromoendoscopy.

6. Unable to obtain informed consent

Trial design

510 participants in 2 patient groups

CADx

Description:

Histopathology prediction by CADx device

Treatment:

Device: CADx

Endoscopist

Description:

Real-time histopathology prediction by expert and non-expert endoscopists