Abstract
The goal of this project is to obtain quantitative assessments of intraretinal features by determining the structural and optical properties of retinal tissue with optical coherence tomography. To accomplish this goal, structural and optical properties, in addition to thickness measurements, were extracted from OCT-based images and were used for the discrimination of DM eyes with and without DR from healthy eyes. First, structural parameters including the thickness, fractal dimension, energy, entropy, correlation and contrast were evaluated for each intraretinal layer using various image processing techniques such as speckle noise removal, retinal segmentation and blood vessel shadow removal. In addition, optical properties such as the mean reflectance, total reflectance, layer index and scattering coefficients were calculated. Specifically, in this dissertation, the main contribution from the biomedical engineering perspective was the development of single and multiple-scattering models using information from different cellular layers of the retina. These models were implemented to extract scattering coefficients from OCT images of healthy and diseased eyes. There is little published work addressing the optical properties of retinal tissue, and what research there is uses a scattering model that considers the retinal tissue as a whole without taking into account its multi-layer structure. In contrast, our scattering models allow us to obtain the scattering coefficient for each intraretinal layer and better explore the optical properties of the retinal tissue. Second, statistical analyses including ANOVA followed by Newman-Keuls post-hoc analysis and receiver operating characteristic analysis were performed on the structural and optical parameters between study groups to determine the diagnostic ability of each structural and optical characteristic to differentiate the diabetic eyes with and without MDR from healthy eyes. Based on the statistical analysis, the capacity of each structural and optical parameter to aid in diagnosis can be determined. The results indicated that the methodology shows greater capability for differentiating diabetic eyes with and without MDR from healthy eyes than the standard commercial OCT device. Moreover, the structural and optical parameters that were best able to discriminate diabetic eyes from healthy eyes were evaluated and validated by artificial neural networks with Bayesian radial basis function. Finally, an additional evaluation in a small group of patients with multiple sclerosis, which is another type of retinal pathology manifesting as retinal neurodegeneration, was evaluated based on the developed methodology. Our results have demonstrated that our methodology would yield better insight into the macular pathology and therefore should play an important role in the future of the diagnosis and follow-up of neurological diseases.