U-Net for B-Scan Segmentation: Retinal Layer Detection in OCT Imaging

Optical coherence tomography is an non-invasive imaging modality in ophthalmology that provides high-resolution tissue morphological images of the retina. Optimal layer division of the intra-retinal is desirable for the timely detection, prognosis, and treatment of retinal diseases, including AMD and DME. This work seeks to address some of the issues including noise, low contrast between the layers of the retina, and differences in human anatomy. The goal is to design a method using principles of digital image processing and such deep learning network as 2D U-Net to provide accurate segmentation of the retinal layers. The suggested approach will promote high accuracy, speed and the unconditioned toughness to various databases. Thus, the achieved objective will give clinicians the reliable means to determine the structural alterations in the human retina and improve the possibilities to diagnose diseases at early stages, as well as to develop the individual approach to the treatment. Additionally, the segmentation results will be beneficial to clinical research and application as the potential detailed measurement of the human retina for disease progression and treatment assessment