Novel techniques to improve pathology detection and resection within the lower gastrointestinal tract
File(s)
Author(s)
Nazarian, Scarlet
Type
Thesis
Abstract
Colorectal cancer (CRC) remains a major global health burden. Despite established screening tools such as colonoscopy and advanced imaging, many pre-cancerous or cancerous lesions are still missed or inadequately removed. Recent advances in machine learning (ML) have shown promise in improving lesion detection and characterisation in the lower gastrointestinal (GI) tract, with the aim of preventing post-colonoscopy CRC and reducing residual disease. This thesis explores novel screening tools and intra-operative techniques to enhance pathology detection and resection accuracy in the lower GI tract.
The effectiveness of colonoscopy is limited by operator dependency, leading to missed polyps. As the adenoma detection rate (ADR) inversely correlates with CRC risk, strategies to enhance colonoscopy quality were evaluated. The integration of artificial intelligence (AI) into endoscopy offers improved real-time diagnostic accuracy and procedural outcomes.
A systematic review and meta-analysis demonstrated high diagnostic accuracy of ML tools for colorectal polyp detection and characterisation, with potential to improve ADR and reduce CRC incidence. A second meta-analysis confirmed ML’s ability to identify commonly missed lesions, including diminutive and non-pedunculated adenomas. A cross-sectional study of endoscopists revealed overall positive attitudes toward AI adoption, alongside key perceived challenges to clinical integration.
In surgical management, achieving clear margins is critical to reducing recurrence and improving survival. To address limitations of current intra-operative assessment tools, an ex-vivo study using diffuse reflectance spectroscopy (DRS) with ML classifiers successfully differentiated normal and tumour colonic tissue. This was followed by an in-vivo feasibility study confirming the DRS system’s diagnostic accuracy and ergonomic suitability for real-time intra-operative use.
This thesis highlights the potential of AI and optical technologies to transform CRC detection and resection, offering promising avenues to enhance precision and improve patient outcomes.
The effectiveness of colonoscopy is limited by operator dependency, leading to missed polyps. As the adenoma detection rate (ADR) inversely correlates with CRC risk, strategies to enhance colonoscopy quality were evaluated. The integration of artificial intelligence (AI) into endoscopy offers improved real-time diagnostic accuracy and procedural outcomes.
A systematic review and meta-analysis demonstrated high diagnostic accuracy of ML tools for colorectal polyp detection and characterisation, with potential to improve ADR and reduce CRC incidence. A second meta-analysis confirmed ML’s ability to identify commonly missed lesions, including diminutive and non-pedunculated adenomas. A cross-sectional study of endoscopists revealed overall positive attitudes toward AI adoption, alongside key perceived challenges to clinical integration.
In surgical management, achieving clear margins is critical to reducing recurrence and improving survival. To address limitations of current intra-operative assessment tools, an ex-vivo study using diffuse reflectance spectroscopy (DRS) with ML classifiers successfully differentiated normal and tumour colonic tissue. This was followed by an in-vivo feasibility study confirming the DRS system’s diagnostic accuracy and ergonomic suitability for real-time intra-operative use.
This thesis highlights the potential of AI and optical technologies to transform CRC detection and resection, offering promising avenues to enhance precision and improve patient outcomes.
Version
Open Access
Date Issued
2025-03-30
Date Awarded
01/12/2025
License URL
Advisor
Patel, Nisha
Peters, Christopher
Elson, Daniel
Darzi, Ara
Publisher Department
Department of Surgery & Cancer
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)