TY  - UNPB
A1  - Ahmad, Omer F
M1  - Doctoral
PB  - UCL (University College London)
UR  - https://discovery.ucl.ac.uk/id/eprint/10187421/
N2  - INTRODUCTION: 

There have been rapid advances in artificial intelligence (AI) applied to colonoscopy recently, particularly in tasks such as computer aided polyp detection (CADe). However, there has been limited clinical adoption. The aims of this thesis were to develop a CADe system to improve clinical value and identify major barriers to implementation. 


METHODS: 
A CADe algorithm was developed and evaluated on a high-risk dataset enriched with subtle lesions including a high proportion of advanced lesions. Performance was compared with endoscopists in a video study. An eye-tracking study was performed to investigate perceptual errors. An online survey was used to evaluate endoscopist opinions on different aspects of CADe designs for integration into clinical workflow. A video study was also performed to evaluate inter-observer variation in the perception of simulated CADe false positives. An international study was conducted to prioritise research priorities for the implementation of AI in colonoscopy using a modified Delphi method. 

RESULTS: 
The CADe algorithm demonstrated a high sensitivity for the detection of flat neoplasia, sessile serrated lesions and advanced polyps in an enriched video dataset. The algorithm detected significantly more subtle polyps than endoscopists. Eye tracking studies demonstrated that cognitive errors accounted for the majority of perceptual errors rather than gaze errors. The online survey demonstrated that there was a significant difference in the perception of visual markers, simulated false positives and their interference with regular workflow. The top ten research priorities for implementation were grouped into five major themes including clinical trial design, technological developments, integration into endoscopy workflow, data and regulatory approvals. 


CONCLUSIONS:  
Novel findings and insights into key areas where the development of AI in colonoscopy could be improved to provide further value have been identified. Current major barriers to AI implementation in routine practice are prioritised for future research.
ID  - discovery10187421
N1  - Copyright © The Author 2024.  Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/).  Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms.  Access may initially be restricted at the author?s request.
AV  - public
Y1  - 2024/02/28/
EP  - 318
TI  - Development and Implementation of Artificial Intelligence in Colonoscopy
ER  -