Please use this identifier to cite or link to this item:
|Title:||Advanced image processing techniques for detection and quantification of drusen|
|Authors:||Mora, André Teixeira Bento Damas|
|Publisher:||Faculdade de Ciências e Tecnologia|
|Abstract:||Drusen are common features in the ageing macula, caused by accumulation of extracellular materials beneath the retinal surface, visible in retinal fundus images as yellow spots. In the ophthalmologists’ opinion, the evaluation of the total drusen area, in a sequence of images taken during a treatment, will help to understand the disease progression and effectiveness. However, this evaluation is fastidious and difficult to reproduce when performed manually. A literature review on automated drusen detection showed that the works already published were limited to techniques of either adaptive or global thresholds which showed a tendency to produce a significant number of false positives. The purpose for this work was to propose an alternative method to automatically quantify drusen using advanced digital image processing techniques. This methodology is based on a detection and modelling algorithm to automatically quantify drusen. It includes an image pre-processing step to correct the uneven illumination by using smoothing splines fitting and to normalize the contrast. To quantify drusen a detection and modelling algorithm is adopted. The detection uses a new gradient based segmentation algorithm that isolates drusen and provides basic drusen characterization to the modelling stage. These are then fitted by Gaussian functions, to produce a model of the image, which is used to compute the affected areas. To validate the methodology, two software applications, one for semi-automated (MD3RI) and other for automated detection of drusen (AD3RI), were implemented. The first was developed for Ophthalmologists to manually analyse and mark drusen deposits, while the other implemented algorithms for automatic drusen quantification.Four studies to assess the methodology accuracy involving twelve specialists have taken place. These compared the automated method to the specialists and evaluated its repeatability. The studies were analysed regarding several indicators, which were based on the total affected area and on a pixel-to-pixel analysis. Due to the high variability among the graders involved in the first study, a new evaluation method, the Weighed Matching Analysis, was developed to improve the pixel-to-pixel analysis by using the statistical significance of the observations to differentiate positive and negative pixels. From the results of these studies it was concluded that the methodology proposed is capable to automatically measure drusen in an accurate and reproducible process. Also, the thesis proposes new image processing algorithms, for image pre-processing, image segmentation,image modelling and images comparison, which are also applicable to other image processing fields.|
|Description:||Dissertation presented to obtain the degree of Doctor of Philosophy in Electrical Engineering, speciality on Perceptional Systems, by the Universidade Nova de Lisboa, Faculty of Sciences and Technology|
|Appears in Collections:||FCT: DEE - Teses de Doutoramento|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.