Fadhel Hamdan Jaafar, Hussain
(2012)
Retinal image processing for automated detection and grading of diabetic retinopathy.
PhD thesis, University of Liverpool.
|
Text
569454.pdf - Unspecified Download (22MB) | Preview |
Abstract
The main eye condition associated with diabetes is called diabetic retinopathy and is, the main cause of blindness. The earliest signs of this disease include damage to retinal blood vessels and then the formation of lesions such as exudates and red spots. Such lesions are normally detected manually by clinicians in intensive and time-consuming processes. Computer:-_aided detection and grading of such conditions could facilitate an immediate and accurate Criagnosis. Whilst some progress has been made to detect these diseases, there is no complete system for automated detection and grading of diabetic retinopathy and this is hindering the development of automated methods to support assessment of diabetic eye disease. The aim of this work is to develop computer algorithms that can be used in the medical screening system for evaluating the condition of the retina leading to successful treatment. This work comprises five stages: 1) image pre-processing, 2) retinal structure extraction; 3) hard exudate detection, 4) red lesion detection and 5) grading of diabetic retinopathy. The aim of image pre-processing is to prepare the image with better quality where shade correction using morphological processes and contrast enhancement using fuzzy logic-based method are applied to the image. In the retinal structure extraction, multi-scale morphological technique and classification procedure are proposed for blood vessel detection. Vasculature loop-based method for the optic disc localisation is proposed, while for fovea localisation, a method based on its features and geometric relationships with the other retinal structures is developed. These methods have the advantage of lower computational complexity and competitive performance compared to the existing related methods. A novel coarse to fine strategy is proposed to detect hard exudates, where a local variation operator is used to calculate the standard deviation around each pixel followed by automated thresholding, morphological operations, and classification to segment coarse hard exudates. To fine-tune the result of coarse hard exudates, two region-based segmentation techniques are investigated to detect fine hard exudates. The significance of this method is manifested by its superior performance, lower computational complexity (compared to the current state of the art) and the ability to deal with a variety of image qualities. A novel red lesion detection method is proposed using mathematical morphology to segment candidate red lesions followed by refining them from traces of retinal structures and then a classification based on red lesion features is used to detect red lesions with high degree of discrimination between genuine red lesions and artifacts and as a result its detection performance has proved to be favourable. Grading of diabetic retinopathy is a very important stage after the detection of retinal lesions to evaluate their severity and to decide appropriate treatment. The most reliable medical approaches to diabetic retinopathy grading were investigated to build a novel computer-aided model for automated grading based on the clinical criteria and results of the earlier lesion segmentation. This model quantifies the nature, extent and spatial distribution of all the detected features and provides a clinical grading assessment. This is among the first of such models published and as such the novelty is considered to be one of the main contributions of this thesis.
| Item Type: | Thesis (PhD) |
|---|---|
| Depositing User: | Symplectic Admin |
| Date Deposited: | 19 Oct 2023 17:54 |
| Last Modified: | 19 Oct 2023 18:01 |
| DOI: | 10.17638/03174262 |
| Copyright Statement: | Copyright © and Moral Rights for this thesis and any accompanying data (where applicable) are retained by the author and/or other copyright owners. A copy can be downloaded for personal non-commercial research or study, without prior permission or charge |
| URI: | https://livrepository.liverpool.ac.uk/id/eprint/3174262 |
Dimensions
Dimensions
