Optical coherence tomography (OCT) is becoming an increasingly important modality for the noninvasive assessment of a variety of ocular diseases such as glaucoma, diabetic macular edema, and age-related macular degeneration. Even though individual layers of the retina are visible on OCT images, current commercial quantitative assessment is limited to measuring the thickness of only one layer. Because each intraretinal layer may be affected differently by disease, an intraretinal layer segmentation approach is needed to enable quantification of individual layer properties, such as thickness or texture. Furthermore, with the latest generation of OCT scanner systems producing true volumetric image data, processing these images using 3-D methods is important for maximal extraction of image information. In this thesis, an optimal 3-D graph search approach for the intraretinal layer segmentation of OCT images is presented. It is built upon the optimal 3-D multiple surface graph-theoretic approach presented by Li et al. (K. Li, X. Wu, D. Z. Chen, and M. Sonka, "Optimal surface segmentation in volumetric images - a graph-theoretic approach," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 1, pp. 119-134, 2006). In this method, multiple surfaces can be found simultaneously by transforming the 3-D segmentation problem into finding a minimum-cost closed set in a corresponding vertex-weighted geometric graph. However, the original formulation of this approach did not incorporate varying feasibility constraints or true regional information, two extensions that would aid in the intraretinal layer segmentation of OCT images. Thus, the major contributions of this thesis include: 1) extending the optimal 3-D graph-theoretic segmentation approach to allow for the incorporation of varying feasibility constraints and regional information, 2) developing a method for learning varying constraints and cost functions from examples for use in the approach, 3) developing and validating a method for the 3-D segmentation of intraretinal layers in both time-domain and spectral-domain OCT images (achieving error rates comparable to that of multiple human experts), and 4) analyzing layer thickness properties in normal subjects and in patients with anterior ischemic optic neuropathy (AION).

Optical coherence tomography (OCT) is a safe and non-invasive imaging technique providing high axial resolution. A spectral-domain OCT scanner capable of acquiring volumetric data of the retina is becoming an increasingly important modality in ophthalmology for the diagnosis and management of a variety of retinal diseases such as glaucoma, diabetic retinopathy and age related macular degeneration (AMD) which are major causes of a loss of vision. To analyze and track these ocular diseases, developments of the automated methods for detecting intraretinal layers, optic discs and retinal blood vessels from spectral-domain OCT scans are highly required recently. The major contributions of this thesis include: 1) developing a fast method that can automatically segment ten intraretinal layers in the spectral-domain macular OCT scan for the layer thickness analysis, 2) developing a method that can automatically segment the optic disc cup and neuroretinal rim in the spectral-domain OCT scan centered at the optic nerve head (ONH) to measure the cup-to-disc ratio, an important structural indicator for the progression of glaucoma, and 3) developing a method that can automatically segment the 3-D retinal blood vessels in the spectral-domain ONH-centered OCT scan to extract 3-D features of the vessels for the diagnosis of retinal vascular diseases.