This thesis considers two distinct but related topics concerned with the analysis of images of the ophthalmic fundus. The first involved the differentiation of automatically detected retinal microaneurysms from other spurious objects. A review of the current literature indicated that there had been no rigorous comparison of automated methods of classification for this type of ophthalmological task. Three classification techniques were investigated; a rule based system, linear discriminant analysis and a learning vector quantizer artificial neural network. Each classifier was trained and tested on the same pair of datasets, and the results analysed using receiver operating characteristic curves. It was found for this application that the rule based system performed marginally better than the linear discriminant analysis approach and both were superior to the neural network method. Whilst the improved performance of the rule based system may, in this clinical diagnostic situation, justify its higher development effort, the simplicity and transparency of the statistical method had much to commend it. The second investigation explored methods of completely segmenting the retinal vessel structure from the fundus image. Although a number of studies have considered different parts of the problem, few integrated solutions have been proposed. Each of the steps required were identified and analysed with the objective of integrating them into a fully automated segmentation and analysis system. Some of the problems were successfully resolved but others were either not completely resolved or were shown to be intractable given the current visual description of the problem.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:340629 |
| Date | January 2000 |
| Creators | Frame, Allan |
| Publisher | University of Aberdeen |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://digitool.abdn.ac.uk/R?func=search-advanced-go&find_code1=WSN&request1=AAIU602026 |
Page generated in 0.0014 seconds