Automatic Identification of Pathology-Distorted Retinal Layer Boundaries Using SD-OCT Imaging

Abstract

Identification of these boundaries can help to determine the loss of neuroretinal cells or layers and the presence of retinal pathology, which can be used as features for the automatic determination of the stages of retinal diseases.

Automatic Segmentation of Macular Holes in Optical Coherence Tomography Images: A review

“…The review indicates many types of preprocessing techniques: dilation and erosion [5], median filter [27,107,111], gaussian filter [11,100,101], wiener filter [11,81,99], binary image [26,100], gradient image [26,114], anisotropic difusion filter [5,97,99], image aligment [19,98,103], attenuation coefficient [101], enhanced contrast [1,105], image flattenig [106,114], resize the image [1,107], edge flow [112], sparse filter [112], normalization [81], green channel [81], greyscale [1], morphological operations [1] and others. In Figure 2 is shown the improvements of preprocessing applied to OCT image.…”

“…Some segmentation techniques applied to OCT images are: snakes or active contour, it is a method used to circumvent objects in an image [105]; graph search, this technique starts from a starting point to an ending point, repeating the process, trying to find the best path [5,26,27]; Dijkstra shortest path search, it is an algorithm that seeks to find the shortest path between two nodes [26,99,107]; local gaussian distribution fitting (LGDF), local entropy defines gray level weights [11]; curvature-based surface cutting, it can be flattened into the plane with low metric distortion [11]; ReLayNet, it is an architecture used to segment retinal layers [80,117]; speed-up robust features (SURF), it is a method used in tasks like object recognition, classification etc [115]; adjusted mean arc length (AMAL), it enables to pass the load limit and turning points, and consequently to follow the post-critical equilibrium trajectories [26]; gradient vector flow (GVF), it is an algorithm that locates object edges [96,118,119].…”

“…Other techniques are: multi-scale spatial pyramid (MSSP), it captures the geometry of retina at multiple scales [19]; geodesic distance method (GDM), it can locate pixels in boundaries of layers [43]; convolutional neural network (CNN), it is pooling, which is a non-linear down-sampling [28,108,120]; dynamic programming (DP), it is a method that divide problems and solves each one separately [65,100,102]; canny edge detection, it is an algorithm that detects edges [97,99,104]; markov gibbs random field (MGRF), it allows to derive a global texture description by specifying local properties of textures [98]; loosely coupled level sets (LCLS), it is a technique that uses local intensity variations to segment layers [101]; structure tensor, it utilizes the gradient of a point with neighborhood to get directions of segmentation [102,104]; Randon Forest (RF), that train the data to estimate boundary probabilities [111,121,122]; and OTSU algorithm, it is used to perform automatic image thresholding [102,107,112]. In Figure 3 is shown an example of segmentation approach applied to an OCT image, in which the top boundary of the ILM and RPE layers are highlighted.…”

Fused Detection of Retinal Biomarkers in OCT Volumes

“…To support clinicians with OCT-based diagnosis, numerous automated methods have attempted to segment and identify specific biomarkers from OCT scans. For instance, retinal layer [2,3,4] and fluid [4] segmentation, as well as drusen detection [5] have previously been proposed. While these methods perform well, they are limited in the number of biomarkers they consider at a time and often use pixel-wise annotations to train supervised machine learning frameworks.…”

A Deep Learning Framework for Segmentation of Retinal Layers from OCT Images

“…Automatic segmentation of 3 layers relevant to age related macular generation and geographic atrophy was proposed [2] by adapting the edge weights used in graph theory and dynamic programming based framework [1]. More recently, information such as slope similarity and nonassociativity of layers as edge weight have been explored to handle pathologies [9].…”