Clinical vAliDation of ARTificial Intelligence in POlyp Detection (CAD-ARTIPOD)

Nuovo Regina Margherita Hospital, Rome, Italy, Krankenhaus Barmherzige Brüder, Regensburg, Germany, Centre Hospitalier Universitaire de Nantes, Nantes, France, Centrum Onkologii-Instytut im. Marii Skłodowskiej-Curie, Warschau, Poland, Spire Portsmouth Hospital, Portsmouth, United Kingdom, University Medical Center, Amsterdam, The Netherlands, University Hospitals Ghent, Ghent, Belgium

This study is an open label, unblinded, non-randomized interventional study, comparing the investigational artificial intelligence tool with the current "gold standard": Data acquisition will be obtained during one scheduled colonoscopic procedure by a trained endoscopist. During insertion, no action will be taken, colonoscopy is performed following the standard of care. Once withdrawal is started, a second observer (not a trained endoscopist but person trained in polyp recognition) will start the bedside Artificial intelligence (AI) tool, connected to the endoscope's tower, for detection. This second observer is trained in assessing endoscopic images to define the AI tool's outcome. Due to the second observer watching the separate AI screen, the endoscopist is blinded of the AI outcome. When a detection is made by the AI system that is not recognized by the endoscopist, the endoscopist will be asked to relocate that same detection and to reassess the lesion and the possible need of therapeutic action. All detections are separately counted and categorized by the second observer. All polyp detections will be removed following standard of care for histological assessment. The entire colonoscopic procedure is recorded via a separate linked video-recorder.

This is an investigator-initiated non-randomized prospective interventional trial to validate the performance of a novel state-of-the-art computer-aided detection (CADe) tool for colorectal polyp detection implemented as second observer during routine diagnostic colonoscopy and to evaluate its feasibility in daily endoscopy. Consecutive patients referred for a screening, surveillance or diagnostic colonoscopy will be included. Patients will undergo a standard colonoscopy performed by a trained endoscopist. A second observer, who is not a trained endoscopist, will follow the procedure on a bedside AI-tool to count the number of detections made by the AI system and categorize the results into positive or negative results as follows (1) true positive, (2) false negative or (3) false positive. In case of a detection of the AI-system that was not seen by the endoscopist or unclear to the second observer, the second observer will ask to re-evaluate the indicated region to determine whether after second look the endoscopist has to take extra action. The entire procedure will be recorded. There are no additional risks specific to the use of the AI tool to be taken into account. General risk of colonoscopy (i.e.: perforation, bleeding or post-polypectomy syndrome) could occur with the same frequency as that of a colonoscopy without the use of this AI tool. All patients will receive a standard of care protocol during their colonoscopy. The AI system can only have a beneficial outcome for the patient, a better polyp detection, as it has shown to be non-inferior in terms of accuracy when compared to high detecting endoscopist in our pilot trial

This is an investigator-initiated non-randomized prospective interventional trial to validate the performance of a novel state-of-the-art computer-aided detection (CADe) tool for colorectal polyp detection implemented as second observer during routine diagnostic colonoscopy and to evaluate its feasibility in daily endoscopy.

Only one arm in this study. Every patient who is eligible for this study and is included, after informed consent, will receive a standard colonoscopy combined with real-time AI video analysis

Patients will undergo a standard colonoscopy performed by a trained endoscopist. A second observer, who is not a trained endoscopist, will follow the procedure on a bedside AI-tool to count the number of detections made by the AI system and categorize the results into positive or negative results as follows (1) true positive, (2) false negative or (3) false positive.

Total polyp detection during single pass colonoscopy by the artificial intelligence tool in comparison to polyp detection by the endoscopist with endoscopic diagnosis as a gold standard

Total polyp detection during single pass colonoscopy by the artificial intelligence tool in comparison to polyp detection by the endoscopist with histological diagnosis as a gold standard.

The number of extra detected polyps by artificial intelligence with the endoscopic diagnosis as a gold standard.

The number of extra detected polyps by artificial intelligence with the histological diagnosis as a gold standard

Correlation between the Boston Bowel Preparation Score and the number of false positive detections during colonoscopy

Correlation between the endoscopist's historical adenoma detection rate and the number of extra detections and false negative detections by the artificial intelligence system.

Correlation between the Paris classification and the number of false negatives and additional detections.

Inclusion Criteria: Age ≥40 years Referral for screening, surveillance or diagnostic colonoscopy Able to give informed consent by the patient or by a legal representative Exclusion criteria for study inclusion <40 years old Referral for a therapeutic colonoscopy Known Lynch syndrome or Familial Adenomatous Polyposis syndrome Any contraindication for colonoscopy or biopsies of the colon Uncontrolled coagulopathy Confirmed diagnosis of inflammatory bowel disease prior to the scheduled colonoscopy Short bowel or ileostomy Pregnancy Exclusion criteria for study analysis Colonic inflammation of > 30cm during colonoscopy Incomplete colonoscopy for any reason Incomplete recording or technical failure of the artificial intelligence system

We do not plan to make individual participant data available. We might share a overview of anonymized data with the collaborating institutions.