This thesis presents an automated framework for quantitative analysis of fluorescein angiographies of the human retina. Such framework represents the core of a computer-aided system, which can assist NHS clinicians in early diagnosis of macular degeneration (AMD). The presented methodology aims to demonstrate the technical feasibility of automated extraction of retinal blood flow parameters, and results in a step forward in the development of an automated computer vision system for quantitative analysis of fluorescein angiograms to assist NHS clinicians in early diagnosis of AMD. The approach commences by segmenting the anatomic constituent parts of the ocular fundus, i.e. the optic disc (OD), the fovea and the vascular network tree. The OD/fovea are simultaneously detected by combining luminance information and geometric information from the major blood vessels; information regarding OD/fovea is then used for delineating the macula. Meanwhile, to segment the retinal vasculature, three independent approaches are implemented. These approaches use information about maximum curvature in both image- and frequency-domains. Such information is combined, firstly, using a supervised linear classifier. Secondly, utilising a committee of local experts, where each expert is represented by an artificial neural network. A fuzzy clustering algorithm is used for expert selection based on the specific input pattern. The output of the system is determined through a winner-takes-all rule. And finally. by considering a tracing algorithm that follows vessel centrelines and walls using a set of rules based upon information of maximum curvature and symmetry. Following segmentation, the extracted vasculature is utilised as input features for a multi modal registration algorithm, which has its fundamentals on the Fourier transform and a parametric estimation based on the gradient of the quadratic error function and least squares computation. Once subsequent frames of the angiogram have been aligned, anatomic, morphologic and sequential analyses are carried out. Special attention is given to the latter one, which is a methodology for quantitative analysis of retinal haemodynamics. It analyses retinal blood flow based on the estimation of parameters such as mean transit-time (MTT) and vascular volume. The former parameter is estimated using densitometry and analysis of the vascular response; the latter is calculated from the lumen of extracted vessels. The performance of the framework is demonstrated on a comprehensive dataset, which contains images of normal retinas and retinas with pathologies such as wet age-related macular degeneration and branch retinal vein occlusion. Results achieved in certain individual modules overcame serious defects observed in previous methods.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:553699 |
| Date | January 2012 |
| Creators | Vargas Canas, Rubiel |
| Publisher | City University London |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
Page generated in 0.0019 seconds