Developing an Automated Algorithm Detection of Micro Aneurysms in Images of Retina for The Purpose of Early Diagnosis of Diabetic Retinopathy

Author: Ahmed Alrashdi

Alrashdi, Ahmed, 2016 Developing an Automated Algorithm Detection of Micro Aneurysms in Images of Retina for The Purpose of Early Diagnosis of Diabetic Retinopathy , Flinders University, School of Computer Science, Engineering and Mathematics

Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.

Abstract

Diabetes is one of the leading problems not only faced by persons aged 25-74 years but it is spreading in young generations too. Diabetes can cause inability of the eye to focus with the increase of diabetes from moderate to severe conditions. This symptom leads to optometrist’s diagnosis of patients. Several years of increase of Diabetes in a person leads damage of vision and can even cause blindness. These various factors lead towards the study of Diabetic Retinopathy. The extract mechanism which leads to Diabetic Retinopathy is given in various studies, but still there is a need to study and formulate the typical problem and its history.

The motivation for this work includes the provision of detection of retinopathy in its early stage so that the prevention steps could be taken and hence protects the vision of a person suffering from diabetes. Microaneurysms (MAs) are defined as the first stage of retinopathy in which we find the red spots in the superficial layer of retina walls. A model has been proposed as a new, alternative method construct profiles of the image intensity centred on the candidate MA in various directions. The main challenge in detection of MA is to find the difference between the true positives and false positives in such a way that accurate detection is available at the output.

In the past few years a large number of algorithms for candidate MA detection have been developed. A broad range of detecting algorithm follows morphological and non-morphological schemes. However, many of the methods, proposed previously failed to give good performance at the output. Istvan Lazar and Andras Hajdu [2013] proposed a method where candidate MA have been identified. In the paper, the authors then construct profiles of the image intensity centered on the candidate MA in various directions.

A new scheme to make some improvement on this is to model the intensity at candidate MA as the function. In the proposed scheme, for a single direction θ, we use logarithmic regression to fit the data on a line in direction θ to the model to obtain a number λ (θ). This is repeated for several directions θ1, θ2,...,θn. If the candidate MA is really an MA, then all the λ (θi) should be fairly large and about the same size. If the candidate MA is really due to the crossing of two vessel crossings, then the λ(θi) will vary greatly. By taking the Fourier transform of the function λ, the pattern of oscillation becomes rotation invariant. Thus local maxima

are classified as a true MA or not based on the Fourier transform of lambda. One advantage of the Fourier transform is that it shifts invariant and in this case θ invariant. Thus the orientation of any background structure such as vessels is removed. For a true MA, nearly all the energy should appear in the DC component of λ. For a candidate MA that is due to the crossing of two vessels, substantial energy should appear in other Fourier coefficients. Testing which of these two methods works better and finding out how many Fourier coefficients to use is be part of the study.

Keywords: Diabetic Retinopathy, Microaneurysms, morphological , non-morphological, Fourier transform

Subject: Computer Science thesis, Medical Biotechnology thesis, Engineering thesis

Thesis type: Masters
Completed: 2016
School: School of Computer Science, Engineering and Mathematics
Supervisor: Associate Professor Murk Bottema