Phoommanee, Nonpawith; (2025) Development of AI endoscopic techniques to perform objective differential diagnosis of nasal obstruction. Doctoral thesis (Ph.D), UCL (University College London).

Text Phoommanee_10205582_Thesis.pdf Access restricted to UCL open access staff until 1 March 2026. Download (4MB)

Abstract

Automated differential diagnosis systems for nasal obstruction using computer vision can assist ear, nose, and throat specialists in treatment management, preventing unnecessary treatments, and aiding patient expectations. This minimally invasive framework has potential for clinical transition from secondary care to community care settings. A combination of rule-based expert systems and machine learning techniques has been used in medical diagnostic applications as it provides interpretable and transparent frameworks compared to black-box approaches for the diagnostic evaluation of patients with multiple conditions. This builds clinicians' trust in machinelearning-based models. This thesis develops a combined rule-based expert system and machine learning framework, including support vector machine and deep learning, for automated nasal obstruction differential diagnosis from nasal endoscopy videos. This thesis involves three evolutionary stages. The first stage evaluates existing classification systems quantifying nasal obstruction severity from nasal endoscopy videos. The second stage develops a rule-based expert system for differentiating common nasal obstruction conditions: allergic rhinitis, chronic rhinosinusitis with or without nasal polyps, and deviated nasal septum. The third stage develops ensemble machine learning methods to obtain rule-based system inputs from nasal video recordings. Key contributions of this thesis include: 1) collecting and annotating the first anterior nasal endoscopy dataset (NE-UCLH dataset) from 71 participants with/without nasal obstruction; 2) collating existing classification systems quantifying nasal obstruction from anterior rhinoscopy or nasal endoscopy, and validating the selected classification systems with nasal endoscopy videos; 3) quantifying the diagnostic logics of ENT specialists using a decision table; 4) developing and validating the first combined rule-based expert system and machine learning approach for differentiating main nasal obstruction conditions. This standalone imaging framework can potentially enhance current ENT diagnostic frameworks.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item