Retinal layer segmentation from optical coherence tomography images

Abstract

Three out of the four leading eye diseases affect the retina, causing irreversible blindness and various degrees of visual impairment. In the clinic, the effects of these diseases and other cardiovascular disorders are attributed to structural changes in the retinal structures. These changes are evaluated using various imaging techniques such as fundus imaging and optical coherence tomography (OCT). Consequently, the analysis of these images has become vital for diagnosing various ocular diseases in modern ophthalmology. Many computer-aided diagnostic (CAD) methods have been proposed to aid with the analysis due to the complexity of the retinal structures, the tediousness of manual segmentation and variation from different specialists. Besides, the commercially available systems focus on only a few layers of the retina, even though recent researches in the field of ophthalmology and neurology show that each layer might be affected individually. The reasons mentioned earlier urge for efficient intra-retinal layer segmentation methods. However, image artefacts such as speckle noise and inhomogeneity in pathological structures remain a challenge, with negative influence on the performance of segmentation algorithms. This study investigates methods for image analysis, aiming to develop robust algorithms for segmenting retinal OCT images. Hence, this thesis presents four methods for extracting individual layer information from OCT to help with eye screening and management of various eye disorders, including glaucoma, diabetic retinopathy, age-relatedmaculardegeneration, among others. Distinctly, the first method is a comprehensive and fully automated method for annotation of retinal layers in OCT images, which comprises of fuzzy histogram hyperbolization for weight reassignment within adjacency matrices and graph-cut (shortest path) to segment seven (7) layers across eight (8) boundaries of the retina. Second, prior knowledge of the retinal architecture derived from the gradient information is embedded into the level set method to segment seven (7) layers of the retina. This method starts by establishing a region of interest (ROI), and then the refined gradient edges obtained from the ROI are used to initialise a level set function. Then, the understanding of layer topology is used in constraining the evolution process towards the actual layer boundaries.