The purpose of the study is to assess whether the AI characterisation system of the CADDIE device improves the endoscopists accuracy in the optical diagnosis of diminutive colorectal polyps in the bowel during colonoscopy. Participants will either have a colonoscopy with the assistance of the CADDIE device characterisation AI system ("intervention group") or have a colonoscopy in line with routine clinical practice i.e., without the CADDIE device characterisation AI system ("control group"). The randomisation method of this trial will allocate enrolled participants to the "intervention" group and to the "control" group by a technique similar to flipping a coin.

The CADDIE system comprises a computer attached to the output of the endoscopy stack and a screen which shows the normal video output of the colon in real-time. Using artificial intelligence (AI) software, it highlights the characteristics of a polyp to help the endoscopist in their optical diagnosis of the polyp histology. Pressing on a foot pedal, which is connected to the computer, presents the information from the AI software to the user as an overlay on the endoscopy screen in text format.

The CADDIE device is capable of recording the video output of the colon in real-time irrespective of whether the CADDIE's AI system is active.

In the endoscopist recruitment phase, we will identify endoscopists expertise and experience in colonoscopy.

In the study phase of the trial, the endoscopist will be kept the same for the CADDIE and control arm. At the beginning of each procedure, participant information will be entered into an electronic system (Redcap) which will generate a random allocation into one of two arms. The CADDIE arm involves a colonoscopy assisted by the CADDIE AI system (i.e., AI software). The control arm is a colonoscopy without the CADDIE's polyp characterisation AI system (i.e., standard practice). Each endoscopist will have the CADDIE device present during all endoscopy lists and will record all procedures in both CADDIE and control arms. The CADDIE's polyp characterisation system (i.e., AI software) will only be active in the CADDIE arm. The recordings in the control arms will allow an extra post-hoc assessment of how CADDIE might have performed compared to the actual assessment by the endoscopist.

In both arms of the trial, for each polyp the endoscopist detects, they will:

1. Measure the polyp size and capture images of the polyp:

   Polyps are to be measured using an instrument of known size. For polyps measured to be <10mm, the endoscopist will capture a still image of the polyp in both white light and Narrow Band Imaging (NBI) or equivalent imaging modality.

2. Optically diagnose the polyp:

   The endoscopist will optically diagnosis polyps <10mm, stating whether the diagnosis is made with high or low confidence.

3. Resect diminutive polyps:

   Management of rectosigmoid hyperplastic polyps will be standardised. Rectosigmoid polyps that are optically diagnosed as hyperplastic by the endoscopist will be resected/biopsied. If multiple rectal hyperplastic polyps are optically diagnosed by the endoscopist, a maximum of 5 will be resected/biopsied for histopathology.

4. Label the polyp with unique ID:

   Each polyp resected (regardless of size) must be labelled according to instructions of the Specimen Handling SOP. Each unique polyp ID will be logged in chronological order on a per colonoscopy basis and sent for histopathology. Histopathology will be referenced as the ground truth for polyp diagnosis and characterisation.

5. Complete documentation as per standard of care:

   Documentation will be made in the endoscopy report of polyp site, size, morphology, segmental Boston Bowel Preparation Score (BBPS). A photograph will also be taken of the caecum and stored in the endoscopy report.

6. Complete the optical diagnosis case report form (CRF):

Populate the optical diagnosis CRF at the end of the procedure with the polyp ID of each polyp sample (regardless of size), their optical diagnosis of each polyp <10mm in size and whether their optical diagnosis was made with high or low confidence.

In the CADDIE arm, the endoscopist presses the foot pedal once they have frozen an image of the polyp in the endoscopy screen in narrow band imaging (NBI) mode. This will output the CADDIE's characterisation of the polyp. Using this information and the endoscopist's own assessment, the endoscopist will log an optical diagnosis of the polyp and the confidence of their diagnosis (high or low). In the control arm, the endoscopist optically diagnoses the polyps without CADDIE's polyp characterisation function. The optical diagnosis will be assessed against the polyp histopathology which will be referred as the ground truth. Post-hoc analysis of the recorded footage will allow for comparison with the CADDIE generated polyp characterisation.

Safety of the device will be assessed through monitoring of adverse events. Acceptability will be assessed through a qualitative questionnaire for the endoscopist, endoscopy nurse and participant in addition to measuring endoscopy markers such as caecal intubation time, procedural time, withdrawal time.

Inclusion Criteria:

• Participants scheduled to undergo a screening, surveillance or symptomatic colonoscopy with an endoscopist participating in the study phase of the trial.
• Male and female participants aged 18 years or older at the time of informed consent.
• Participants able to comprehend, sign and date the written informed consent document to participate in the study.

Exclusion Criteria:

• Emergency and/or inpatient colonoscopies

• Participants with inflammatory bowel disease (IBD)

• Participants with current Colorectal Cancer (CRC)

• Participants with a contraindication for biopsy or polypectomy. These include:

  • Participants who have not withheld medications pre-disposing to bleeding at time of colonoscopy as per local site /national guidelines.
  • Participants with a history of haemostasis disorders (haemostasis disorders will include but will not be limited to: participants with haemophilia or other congenitally acquired clotting factor deficiencies, participants with cirrhosis with coagulopathy, participants known to have thrombocytopenia (<80,000 platelet/ul) and individuals with Von Willebrand's disease or other known platelet malfunction disorders)

• Participant is enrolled in another research study with an investigational medicinal product (IMP) or non-IMP that pre-disposes them to bleeding

Participants scheduled to undergo a screening, surveillance or symptomatic colonoscopy at UCLH

Patients will be asked informed consent with the possibility to share the data

Data will be made available from 1 year after the trial findings are published. This is available for up to 5 years after study completion.

We will make provision for and consider data sharing requests from bona fide researchers who must abide by the following principles: data will be collected with an high level of quality assurance, data will be held securely with appropriate documentation, data will not be put into the public domain or otherwise shared without explicit ethical review or legal obligation, and they will aim to use any data generated to the maximum public good. These datasets are governed by data usage policies specified by the data controller (UCL). If a conflict exists that severely restricts the analyses that can be undertaken, we would endeavour to support outside researchers by hosting them as visiting workers in our team so that they can access the data. We are committed to complying with the Wellcome Trust Data Sharing Policy. Applications are subject to review by Laurence Lovat and Rawen Kader and should be emailed to l.lovat@ucl.ac.uk