Deformation Measurements of the Retinal Nerve Fiber Layer by Automatic Segmentation on High Frequency Optical Coherent Tomography

Abstract

Purpose: Optical Coherent Tomography (OCT) has become the standard clinical and research tool for non-invasive cross-sectional imaging of the posterior layers of the eye. It provides essential information about the eye’s anatomy and physiology, and changes associated with various eye diseases. Beyond clinical use, the vast majority of research projects based on OCT rely on static images made up of hundreds of averaged images, rather than dynamic studies. We have previously demonstrated that peripapillary retinal tissue is deformed during cardiac pulsations, and that such deformation is different in normal and glaucoma patients (Singh et al IOVS 53:7819 (2012). In order to be able to systematically perform such examination, we present a fully automated image analysis algorithm to detect the Retinal Nerve Fiber Layer (RNFL) and assess the deformation of the tissue in video rate FD-OCT images.

Methods: Before the segmentation, images are pre-processed with morphological operations, median filtering and intensity thresholds to separate the outer layers of the retina from the vitreous and inner layers; pixel labeling is then used to remove spurious pixels near the vitreous interface. The segmentation algorithm seeks to trace the surface of the peripapillary retina and Optic Nerve Head (ONH) by selecting the bright anterior foreground pixel per A-scan on the image and the resulting contour is smoothed using a Savitzky-Golay filter. From the obtained retinal surfaces, tissue deformation is quantified in two orthogonal directions: vertical, from peripapillary retina to prelaminar tissue, and horizontal, neuroretinal rim.

Results: The method was tested on 32 normal adult eyes recording OCT scans of the ONH for 20 seconds at 19.5Hz (~400 images per recording in real time). Detection of the RNFL and deformation measurements was successful in 70% of the subjects, segmentation errors on the rest was due to very high vitreous reflectivity, strong eye movements or poor image quality. Based on this data it is possible to estimate the amount of potential stretching of the axons forming the optic nerve.

Conclusions: We developed an automatic algorithm that provides robust detection of the RNFL and ONH allowing tissue deformation measurements. This dynamic measure of retinal movement has important implication for the understanding of retinal and ONH tissue biomechanics.

Keywords: 549 image processing • 550 imaging/image analysis: clinical • 627 optic disc