Use this URL to cite or link to this record in EThOS:
Title: Probabilistic labelling for enhancement of vessel networks applied to retinal images
Author: Paredes Soto, Daniel Alonso
ISNI:       0000 0004 5921 3537
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Occlusive vascular disease affecting arterial circulations is the major and fastest growing health problem worldwide, and underlies common conditions such as heart attack, stroke and peripheral vascular disease. Although vascular diseases may be assessed according to clinical history, screening may be required to evaluate health conditions or courses of treatment. Vasculature in the retina and other organs such as the brain have similar anatomical properties and regulatory mechanisms. Changes in the morphology of retinal vasculature may be associated with vascular-related conditions such as hypertension and stroke. Owing to its high cost-effectiveness, eye fundus photography is often used to study changes in the retinal vasculature. This research proposes a probabilistic pixel labelling method based on analysis of local and global features of the image to enhance the detail of vessel structures. Our approach produces a probability map that could be further used by contextual approaches (e.g. Markov Random Fields) for segmenting vessel networks as future application. We first correct contrast variation due to non-uniform illumination and reflections produced by eye tissue using statistical methods to locally estimate the contrast behind vasculature structures. Our labelling method is based on the Hessian matrix to locally estimate the maximum probability of the principal local curvature—given by eigenvalues—matching an ideal vessel curvature. We defined a realistic model based on imaging physics to produce the ideal vessel curvature governed by the Beer-Lambert Law for estimating the absorption of energy as it is propagated through uniformly filled objects. The local maximum posterior probability—based on Bayes’ rule—was eventually estimated by combining the prior (using the proposed background estimation) and the likelihood produced by Monte Carlo simulations. The proposed method in this research was compared with one of the most popular vessel detectors due to Frangi showing similar results.
Supervisor: Rockett, Peter I. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available