Purchase this article with an account.
Qi Yang, Charles A. Reisman, Kinpui Chan, Rithambara Ramachandran, Ali S. Raza, Donald C. Hood; Automated Segmentation of Retinal Layers in Macular OCT Images of Patients with Retinitis Pigmentosa. Invest. Ophthalmol. Vis. Sci. 2012;53(14):4083.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To provide an automated layer segmentation algorithm for quantifying structural changes of the outer segment and other retinal layers as seen on spectral domain optical coherence tomography scans of patients with retinitis pigmentosa (RP).
The current algorithm employs dual-gradient information to perform shortest path search [2,3]. Connected component analysis and heuristics have been added to improve the accuracy of intra-retinal boundary detection. The algorithm delineates the inner limiting membrane (ILM), the nerve fiber layer (NFL)/ganglion cells layer (GCL) border, the inner plexiform layer (IPL)/the inner nuclear layer (INL) border, INL/outer plexiform layer (OPL) border, and three other outer retinal boundaries in three-dimensional (3D) volume scans. To test the accuracy, the manually segmented border positions were compared to the automatic segmentation results of 88 B-scans acquired from 8 RP macular 3D scans (3D OCT-1000, Topcon Corp., Tokyo, Japan). To test the repeatability, the macular 3D scans of 13 eyes (each with two repetitions) were segmented and the intra-class correlation coefficients and the mean of standard deviations were calculated.
The absolute mean differences between the manual and automated border positions were less than or about two pixels: ILM 3.57±2.50µm, NFL/GCL 7.47±4.66µm, IPL/INL 6.19±2.26µm, INL/OPL 6.94±4.32µm, IS/OS 3.92±1.15µm, OS/retinal pigment epithelium 3.82±1.59µm, and Bruch’s membrane 4.63±2.26µm. The intra-class correlation coefficients for the seven boundaries ranged from 0.92 to 0.98. The standard deviations of segmented boundaries were between 2.64µm and 4.47µm.
The automated RP segmentation algorithm demonstrated good accuracy and repeatability results for seven retinal boundaries. It provides a fast and objective approach to explore possible structural markers for RP progression.1. Hood DC, et al., Biomed Opt Express 2011. 2. Yang Q, et al., Opt Express 2010. 3. Yang Q, et al., Biomed Opt Express 2011.
This PDF is available to Subscribers Only