Artificial Intelligence (AI) Assisted Real-time Adenoma Detection During Colonoscopies

This is a pragmatic, double-blind, randomized, controlled trial, to assess whether the implementation of CADe (GI genius) significantly affects the ADR in screening and surveillance population. The main hypothesis of this study is that the ADR in the operating room equipped with the GI genius CADe system will be significantly higher than the ADR in the ordinary operating room.

This trial will be conducted in five centers across Canada. All patients who meet the in/exclusion criteria can be enrolled. The patient's personal medical history will be reviewed to verify patient inclusion and exclusion criteria (age, history of CRC or adenoma, comorbid conditions, anticoagulation, etc.). Eligible patients will be randomized (1:1) stratified per center in two arms: Intervention arm: patients will be assigned to undergo colonoscopy in a room equipped with the GI genius CAD system. Control arm: patients will undergo colonoscopy in a room not equipped with the GI genius CADe system. Allocation will be concealed and study participation not know to the endoscopist performing the procedure. This study will be a pragmatic double-blinded: the patients and the endoscopist will be blinded to the allocated group. Data will be collected on case report forms (CRF), after the procedure, from the clinical files and the endoscopy reports. The data will then be deidentified and transferred to an electronic RedCap database in each institution. A research assistant will collect all information and annotations recorded on the patient's medical file during the procedure. In the treatment group (GI genius CADe assisted colonoscopy), the Medtronic-GI genius system provides real-time support to endoscopists to detect polyps of any range of sizes. CADe will provide real-time feedback throughout each colonoscopy procedure and will alert the endoscopists of the presence of a polyp in the endoscopy field by displaying a bounding box on the same screen. In the control group (standard colonoscopy), the participating endoscopists will detect and classify colorectal lesions without using any AI modules. The classification of the polyps will also be based on the NICE and SANO classification systems as performed in the treatment group. After detecting a polyp and performing optical diagnosis for all 1-10 mm polyps, the endoscopists will remove all polyps using standard technique\s. 1-2 mm polyps can be removed using a standard biopsy forceps or polypectomy snares, and polyps 3-10 mm with a standard polypectomy snare (preferably cold snares for all small polypus), 10-20 mm polyps are removed using a polypectomy snare (cold or hot upon endoscopist's decision), and >20 mm polyps using EMR or ESD. All polyps will be resected and sent to the pathology labs of the participating institutions to be evaluated for histology by board-certified pathologists. The histopathology outcomes will be collected and stored in the CRF forms to be used as a reference later.

Allocation will be concealed; the participating endoscopists and the patients will not be informed of the treatment group.

A standard colonoscopy will be performed according to the standard of routine care. The AI system will capture video of the procedure in real time, and provide additional information on the detection of polyps and classification. All optically diagnosed polyps will be removed and sent to the hospital pathology laboratory for histopathological evaluation according to institutional standards.

In the control group (standard colonoscopy), the participating endoscopists will detect and classify colorectal lesions without using any AI modules. All diagnosed polyps will be removed and sent to the hospital pathology laboratory for histopathological evaluation according to institutional standards.

The GI-Genius system could reach a sensitivity of 99.7%, a false-positive rate <1%, and performed real-time analysis 82% faster than the endoscopist. The use of this system resulted in a significant increase in ADR compared with the conventional colonoscopy (54.8% vs. 40.4%) in a screening cohort. Medtronic-GI genius: AI will provide real-time feedback throughout each colonoscopy procedure and alert endoscopists to the presence of a polyp in the endoscopy field by displaying a bounding box on the same screen. Then, endoscopists will first perform optical diagnosis without the aid of the AI using the validated NBI (narrow band imaging) International Colorectal Endoscopic (NICE) and Sano classification systems to distinguish non-neoplastic from neoplastic polyps and express confidence in the prediction of histology during optical diagnosis.

To assess whether the GI genius platform can significantly increase the ADR during live colonoscopies, compared to standard colonoscopy among high-risk patients.The ADR is defined as the proportion of patients undergoing GI genius-assisted/standard colonoscopy in whom superior or equal to 1 histopathologically proven adenoma or carcinoma is identified. We will use the histopathology outcomes of the resected polyps as the ground truth to calculate the ADR.

The probability of an increase in the ADR among the trainees and expert endoscopists when using GI genius AI-assisted colonoscopy compared to standard colonoscopies.

Proximal ADR is defined as the prevalence of patients with at least 1 adenoma detected proximal to the splenic flexure

Sessile serrated lesions (SSL) detection rate as the proportion of patients colonoscopy in whom superior or equal to 1 sessile serrated lesion (SSL) is identified

Mean number of adenomas per colonoscopy (APC) defined as the total number of adenomas divided by the number of colonoscopies performed

Adenomas per positive index colonoscopy (APPC) defined as the total number of adenomas divided by the total of colonoscopies where at least 1 adenoma is detected

Sessile serrated lesion per Colonoscopy (SSLPC), defined as the total number of histologically confirmed sessile serrated lesions resected divided by the total number of colonoscopies

Advanced lesion detection rate (ALDR), defined as the proportion of colonoscopies for which the number of histopathologically proven adenoma >10mm was at least 1.

To evaluate the agreement between the surveillance intervals based on the optical diagnosis of polyps with the GI genius and the pathology-based recommendation according to the current guidelines.

Proportion neoplastic and non-neoplastic polyps (i.e., SSLs, hyperplastic polyps, and high-grade dysplasia and carcinomas (overall and according to size, location, and morphology)

CADe utilization rate (when available in the room). Since we can't directly assess the CADe utilization rate without revealing the ongoing trial to the endoscopists, we will rely on two voluntary self-reported measures and a partial objective measure.

Inclusion Criteria: indication of undergoing a screening, surveillance, or diagnostic colonoscopy Age 45-89 years Exclusion Criteria: Patients undergoing emergency colonoscopy and patients with a known familial polyposis syndrome or a known inflammatory bowel disease