Artificial Intelligence Development for Colorectal Polyp Diagnosis

Development of a Novel Real Time Computer Assisted Colonoscopy Diagnostic Tool for Colorectal Polyps: Lesion Diagnosis and Personalised Patient Management

Accurate classification of growths in the large bowel (polyps) identified during colonoscopy is imperative to inform the risk of colorectal cancer. Reliable identification of the cancer risk of individual polyps helps determine the best treatment option for the detected polyp and determine the appropriate interval requirements for future colonoscopy to check the site of removal and for further polyps elsewhere in the bowel.

Current advanced endoscopic imaging techniques require specialist skills and expertise with an associated long learning curve and increased procedure time. It is for these reasons that despite being introduced in clinical practice, uptake of such techniques is limited and current methods of polyp risk stratification during colonoscopy without Artificial intelligence (AI) is suboptimal. Approximately 25% of bowel polyps that are removed by major surgery are analysed and later proved to be non-cancerous polyps that could have been removed via endoscopy thus avoiding anatomy altering surgery and the associated risks. With accurate polyp diagnosis and risk stratification in real time with AI, such polyps could have been removed non-surgically (endoscopically). Current Computer Assisted Diagnosis (CADx, a form of AI) platforms only differentiate between cancerous and non cancerous polyps which is of limited value in providing a personalised patient risk for colorectal cancer. The development of a multi-class algorithm is of greater complexity than a binary classification and requires larger training and validation datasets. A robust CADx algorithm should also involve global trainable data to minimise the introduction of bias. It is for these reasons that this is a planned international multicentre study.

The Investigators aim to develop a novel AI five class pathology prediction risk prediction tool that provides reliable information to identify cancer risk independent of the endoscopists skill.

These 5 categories are chosen because treatment options differ according to the polyp type and future check colonoscopy guidelines require these categories

Study Overview

• Status: Recruiting
• Conditions: Polyp of Colon; Colorectal Polyp
• Intervention / Treatment: Diagnostic test: Colonoscopy

Detailed Description

The use of artificial intelligence in computer-assisted detection (CADe) to detect polyps (pre-cancerous growths) during colonoscopy is gaining increasing interest and acceptance with multiple devices already in the mainstream market. The Investigator know already from work in other countries that detecting more polyps results in a reduced risk of bowel cancer for the patient having the procedure, in the years following their colonoscopy (ie. pre-cancerous growths were detected and removed). This has formed the basis of national bowel cancer screening programmes. With increased detection of colorectal polyps, there is a growing need to correctly identify the nature of the polyp to inform the risk of colorectal cancer with the polyp detected and also the potential future risk to the patient. Accurate polyp diagnosis is also required to determine the correct mode or removal-whether this does require removal at all (leading to conservation of costs and resources in a challenging current climate), whether endoscopic removal is possible and if so by what procedure, whether surgery is required.

Published data demonstrates that approximately one quarter of surgically removed colorectal polyps with patients undergoing major surgery were benign and therefore major surgery could have been avoided with these polyps removed endoscopically reducing the risk of complication and organ preservation for the patient.

Current polyp diagnosis techniques involve the use and interpretation of specialist dyes and magnification endoscopes which come with gaining expertise expertise with an associated learning curve and increased procedure time. It is for these reasons that despite being introduced in clinical practice, uptake of such techniques is limited and current methods of polyp risk stratification during colonoscopy without AI is suboptimal.

Current polyp diagnosis AI (CADx) algorithms are limited to smaller classification Current CADx platforms differentiate between cancerous and non-cancerous polyps which is of limited value in providing a personalised patient risk for colorectal cancer. The development of a multiclass algorithm is of greater complexity than a binary classification and requires larger training and validation datasets. A robust CADx algorithm should also involve global trainable data to minimise the introduction of bias. It is for these reasons that this is a planned international multicentre study

Prospective collection of data:

This study will be conducted alongside usual patient care, but will require research staff to enter data onto a secure web-based report form (REDCAP database). This means that participants will undergo exactly the same procedure, with no differences and no extra visits or data, than would have otherwise have occurred. Participants will be those patients that have been scheduled to have a colonoscopy for the standard reasons. Patients will be invited in the usual way for colonoscopy.

They may - where possible - be sent the PIS with their appointment letter (up to 6 weeks in advance). On arrival in the endoscopy unit, they will be approached by a member of the research team and given a copy of the PIS to read - up to an hour before their procedure. They will be provided face-to-face information and explanation, prior to written consent to allow their data to be collected in the database. As the study does not require any change or additional procedures, The investigator feel that an initial approach on arrival into the endoscopy unit will provide sufficient, appropriate time to consent, even if the PIS has not been read in advance (although it will be sent if possible). The only additional consideration will be the consent to recording of the video (no patient identifiable data will be transferred as part of this aspect).

Once the colonoscopy has been completed, there will be no additional visits.

• Study Type: Observational
• Enrollment (Estimated): 4000

Contacts and Locations

• Study Contact: Name: Shraddha B Gulati, MBBS PHD MRCP; Phone Number: +442032996044; Email: shraddha.gulati@nhs.net
• Study Contact Backup: Name: Olaolu Olabintan, MBBS MRCP; Phone Number: +447939056819; Email: olaolu.olabintan@nhs.net

Study Locations

• United Kingdom
  • London, United Kingdom, SE5 9RS
    • Recruiting
    • King's College Hospital NHS Foundation Trust
    • Contact: Shraddha B Gulati, MBBS PHD MRCP; Phone Number: +442032996044; Email: shraddha.gulati@nhs.net
    • Contact: Olaolu Olabintan, MBBS MRCP; Phone Number: +447939056819; Email: olaolu.olabintan@nhs.net

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: N/A

• Sampling Method: Non-Probability Sample

Study Population

All adult (over 18 years) presenting for screening or symptomatic colonoscopy

Description

Inclusion Criteria:

- Above 18 years at inclusion Symptomatic or screening colonoscopy

Exclusion Criteria:

• Unable to provide informed consent.
• Colitis Associated Dysplasia
• Polyps at surgical anastomosis sites
• Pregnancy

Study Plan

How is the study designed?

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
To achieve an overall accuracy of 85% for the five-classification lesion prediction algorithm.	Sensitivity and Specificity	24 months
Positive and negative predicted value	Assess the accuracy to the trained device	24 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Interobserver agreement of the endoscopists' prediction of histology of polyps during the annotation process.	We will analyse the calculate the histology agreement between the advance endoscopist	36 months
Sub-analysis of the polyp characteristics focused on different gender and ethnicity.	To assess if the prediction of patient gender, ethnicity, and age is possible with use of the developed CADx model.	36 months
This will be a sub analysis of an AI algorithm that is trained to predict polyp histology using the prospective data cohort.	A qualitative analysis of CADx incorrect diagnoses will also be conducted by a multidisciplinary panel to evaluate potential impact	36 months
Learned effects of AI augmented endoscopy on endoscopist optical diagnosis	We will evaluate if the use of AI during colonoscopy can be a learning tools for endoscopist	36 months

Collaborators and Investigators

• Sponsor: King's College Hospital NHS Trust

Study record dates

Study Major Dates

• Study Start (Actual): May 4, 2024
• Primary Completion (Estimated): June 1, 2025
• Study Completion (Estimated): May 30, 2026

Study Registration Dates

• First Submitted: May 7, 2024
• First Submitted That Met QC Criteria: June 5, 2024
• First Posted (Actual): June 6, 2024

Study Record Updates

• Last Update Posted (Actual): June 6, 2024
• Last Update Submitted That Met QC Criteria: June 5, 2024
• Last Verified: June 1, 2024

More Information

Terms related to this study

• Keywords: Colorectal cancer; Artificial Intelligence; Machine learning; Colorectal polyp; Computer assisted diagnosis (CADx)
• Additional Relevant MeSH Terms: Pathological Conditions, Anatomical; Intestinal Polyps; Polyps; Colonic Polyps
• Other Study ID Numbers: 323988

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: UNDECIDED

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No