Computer analysis for registration and change detection of retinal images

Elmuntser, A 2021, Computer analysis for registration and change detection of retinal images , PhD thesis, University of Salford.

[img]
Preview
PDF
Download (6MB) | Preview

Abstract

The current system of retinal screening is manual; It requires repetitive examination of a large number of retinal images by professional optometrists who try to identify the presence of abnormalities. As a result of the manual and repetitive nature of such examination, there is a possibility for error in diagnosis, in particular in the case when the progression of disease is slight. As the sight is an extremely important sense, any tools which can improve the probability of detecting disease could be considered beneficial. Moreover, the early detection of ophthalmic anomalies can prevent the impairment or loss of vision. The study reported in this Thesis investigates computer vision and image processing techniques to analyse retinal images automatically, in particular for diabetic retinopathy disease which causes blindness. This analysis aims to automate registration to detect differences between a pair of images taken at different times. These differences could be the result of disease progression or, occasionally, simply the presence of artefacts. The resulting methods from this study, will be therefore used to build a software tool to aid the diagnosis process undertaken by ophthalmologists. The research also presents a number of algorithms for the enhancement and visualisation of information present within the retinal images, which under normal situations would be invisible to the viewer; For instance, in the case of slight disease progression or in the case of similar levels of contrast between images, making it difficult for the human eye to see or to distinguish any variations. This study also presents a number of developed methods for computer analysis of retinal images. These methods include a colour distance measurement algorithm, detection of bifurcations and their cross points in retina, image registration, and change detection. The overall analysis in this study can be classified to four stages: image enhancement, landmarks detection, registration, and change detection. The study has showed that the methods developed can achieve automatic, efficient, accurate, and robust implementation.

Item Type: Thesis (PhD)
Contributors: Antonacopoulos, A (Supervisor)
Schools: Schools > School of Computing, Science and Engineering
Funders: Libyan Embassy
Depositing User: Adham Elmuntser
Date Deposited: 10 May 2021 08:56
Last Modified: 10 Jun 2021 02:30
URI: http://usir.salford.ac.uk/id/eprint/60093

Actions (login required)

Edit record (repository staff only) Edit record (repository staff only)

Downloads

Downloads per month over past year