Software-based image analysis in ophthalmology

R.S. Koolschijn

Research output: ThesisPd Eng Thesis

Abstract

Diabetes has been shown to lead to neuronal damage in the eye, specifically in the transparent front part of the eye, the cornea, and in the light sensitive layer in the inner part of the eye, the retina. Complications of type II Diabetes manifest themselves in these tissues as changes in the amount and structure of neural tissue. The amount of nerve tissue in the cornea and retina can be quantified by measuring the amount of nerves and the thickness of the neural layers in the cornea and retina, respectively. This quantification can nowadays be performed automatically. However, the automated methods that have been developed for this purpose have never been validated on large datasets and in many cases they have not been used outside of the group that developed the method. There is evidence that the neurodegenerative effects of Diabetes can already be detected at an early stage of the disease. This implies that measurements of neuronal damage, e.g. nerve quantity, could be used for early detection of type II Diabetes.
Nerve quantity can be determined in the cornea using confocal microscopy (CCM), and in the retina using Optical Coherence Tomography (OCT). Therefore there is potential for these two imaging techniques to be used as screening instruments for type II Diabetes.
The Maastricht Study (DMS) is a large-scale epidemiological study focused on learning more about type II Diabetes, including the disease’s cause(s), complications and strategies for prevention. Amongst other measurements, confocal microscopy and OCT were included in the Maastricht Study. The purpose of the present project was to find out whether confocal microscopy and OCT can be used as screening tools for type II Diabetes, by analyzing the data of the Maastricht Study using automated image analysis software platforms. In order to achieve this goal, a framework had to be created to enable the analysis of the data, and in addition the automated analysis methods had to be validated.
As part of this project, three automated analysis methods for corneal images as well as three automated analysis methods for retinal images were validated. Prior to validation the framework for each of these analysis methods was designed. In addition, the measurements themselves were validated by analyzing reproducibility data and performing measurement system analyses. Medium sized datasets (100-200 participants) were used to validate the analysis software platforms. Using the results generated by the different analysis methods, the differences between individuals with and without type II Diabetes were assessed. Finally, the results of the corneal and retinal measurements were combined in a basic multivariate analysis to find out whether combining the two techniques would provide additional value in predicting type II Diabetes.
The results from both the cornea and retina analysis confirmed that there are indeed differences in the quantity of the nerve tissue of individuals with and without type II Diabetes. However, the differences are relatively small (up to 10% of the average), with greatly overlapping distributions for the healthy and type II Diabetes groups. When the results of the corneal and retinal analysis were combined into a multivariate analysis, the two groups became somewhat better distinguishable from each other, but still there is a large overlap between them. In addition, the measurement system analysis showed that the spread of individual measurements is very high compared to the difference between groups, which renders both techniques unsuitable to make definite statements about whether an individual has type II Diabetes or not.
The main conclusion of the project therefore is that confocal microscopy of the cornea and OCT of the retina are not qualified to be used as screening instruments for type II Diabetes on an individual level. The techniques may still be of value for epidemiological research studying the effects of type II Diabetes on a group level. The validation of the automated corneal and retinal analysis method resulted in information on the advantages and disadvantages of each method, and in the end recommendations on which methods are most suitable for image analysis in an epidemiological framework were made.
Original languageEnglish
Awarding Institution
Supervisors/Advisors
  • Cottaar, E.J.E. (Ward), Supervisor
  • Berendschot, Tos T.J.M., External supervisor
  • Schouten, Jan S.A.G., External supervisor, External person
Award date14 Jun 2016
Place of PublicationEindhoven
Publisher
Publication statusPublished - 2016

Bibliographical note

PDEng thesis. - Confidential forever.

Fingerprint Dive into the research topics of 'Software-based image analysis in ophthalmology'. Together they form a unique fingerprint.

Cite this