- Abstract: The assessment of total retinal and choroidal thickness from optical coherence tomography (OCT) images is an important clinical and research task. These thickness measures and their changes represent a fundamental metric extracted from OCT data, since they provide valuable information regarding the eye’s normal anatomy and physiology. Changes in thickness are associated with natural eye development, the progression of various eye diseases, and the development of refractive error. Manual analysis of OCT images is time-consuming and not feasible for large datasets of images. Thus, the development of reliable and accurate methods to automatically segment tissue boundaries in OCT images is fundamental. In this paper, convolutional neural networks (CNNs) are used to calculate the probability of boundary locations in OCT images. The CNN, trained using image patches centred around the boundary of interest, provides a per-layer probability map that marks the most likely predicted location of the boundaries. This map is subsequently traced using a graph-search approach to segment the boundaries. The effect of patch size, network architecture and input image pre-processing on the CNN performance and subsequent layer segmentation is presented. The results are compared with manual image segmentation as well as a fully convolutional network. This work may support the future development of CNN methods for automated OCT boundary segmentation.
- Keywords: onvolutional neural networks, Image segmentation, Optical Coherence Tomography, Human Eye
- Author Affiliation: Queensland University of Technology