Holistically-Nested Edge Detection for AMD Images

Abstract

Macular degeneration associated with age is an eye disorder and is one of the leading causes of near blindness. Environmental and hereditary variables influence age-related macular degeneration (AMD) etiology. OCT (Optical Coherence Tomography) is the first quantitative ocular diagnostic for identifying eye illness. Accurate observation and comprehension of the condition are enhanced using CAD (Computer-Aided Detection) systems. This work aimed to develop a technique for Bruch's membrane, retinal pigment epithelium, and inner limiting membrane (ILM) segmentation in OCT images. The segmentation process combines the method using two deep neural networks, ResU-Net and HED. The conclusions reached were beneficial. The variation of U-Net is ResUNet the residual link is a kind of connection that enables the network to learn the residual rather than the complete mapping between the input and output. This method can improve network learning effectiveness and prevent the disappearing gradients issue. The deep learning model, called holistically nested edge detection (HED), performs image-to-image prediction using fully convolutional neural networks and intensely supervised nets. HED automatically creates extensive hierarchical representations to resolve ambiguity in edge and object boundary recognition (guided by intense supervision on side responses).