A proposal of extracting retinal boundaries for optical coherence tomography using three dimensional active grid

摘要

Three dimensional process could improve the extraction rate in comparison with two dimensional process. We propose a new three dimensional active grid for extraction of retinal boundaries on optical coherence tomography. Ophthalmologist desire a system that can measure retinal thickness automatically for quantitative evaluation on retinal diseases. Conventional method tried extract boundaries of retinal layer i.e. inner limiting membrane (ILM) and retinal pigment epithelium (RPE)). The boundaries in normal retina are extracted almost completely. However, extracted boundaries in disease retina included some errors. Therefore we propose a new method for extracting the boundaries of retinal layer from three-dimensional optical coherence tomography (OCT) images. We use 3D active grid which is dynamic shape model. This paper applied the proposed method to 22 OCT image sets (10 sets of normal retinal images and 12 sets of disease retinal images). Retinal thickness in normal retinal images are obtained by accuracy of 97.54 %, and those in disease retinal images are done by accuracy of 72.32 % compared with the judgment of medical doctor. The proposed method is indicated as an efficient method for extraction of retinal border lines.