University of North Texas
A Computational Framework for Colonoscopy Video Quality Assessment and Structure-Aware Image Analysis
Abstract
dc:descriptionColonoscopy is a critical procedure for the prevention of colorectal cancer, enabling direct visualization of the colonic mucosa and removal of precancerous lesions. However, its effectiveness depends heavily on procedural quality and the interpretation of complex endoscopic data. Current clinical metrics provide indirect assessment of mucosal inspection, while degraded video frames and reliance on appearance-based features limit accurate structural analysis. This dissertation presents a computational framework for automated colonoscopy analysis that addresses both procedural quality assessment and structure-aware image understanding. For procedural analysis, deep learning-based methods are developed to identify procedure type, detect non-informative frames, and estimate inspection quality by quantifying newly observed mucosal regions. A Vision Transformer-based model is used to identify degraded frames by capturing global spatial context, while a quality estimation framework integrates informative frame selection with image registration to quantify newly observed mucosal surface area across video frames. To overcome limitations of appearance-based analysis, this work further introduces a structure-aware approach using depth-based representations. By leveraging monocular depth estimation, a content-based image retrieval framework is developed to compare images based on geometric structure rather than visual appearance. Together, these contributions improve the robustness, interpretability, and clinical relevance of automated colonoscopy analysis and provide a foundation for future intelligent clinical decision support systems.
Degree
thesis:*- Grantor dc:publisher
- University of North Texas
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Rahman, Md Marufi
- Contributors dc:contributor
-
- Fan, Heng
- Bozdag, Serdar
- Oh, Junghwan
- Albert, Mark V.
Subjects
dc:subject × 3Rights
dc:rights- Statement dc:rights
-
- Public
- Rahman, Md Marufi
- Copyright
- Copyright is held by the author, unless otherwise noted. All rights Reserved.
- Language dc:language
- English
Identifiers
dc:identifier.*- Identifier
-
local-cont-no: submission_4983
ark: ark:/67531/metadc2652488