Purchase this article with an account.
H. Zhu, D. P. Crabb, P. G. Schlottmann, D. F. Garway-Heath; 3DSEGMENT: Quantitative Analysis of Retinal Structures From Fourier-Domain Optical Coherence Tomography (FD-OCT) 3D Volume Scans. Invest. Ophthalmol. Vis. Sci. 2008;49(13):3764. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop a novel, fully automated segmentation algorithm (3DSEGMENT) to quantify retinal structures in 3D volume images by FD-OCT.
3DSEGMENT makes use of reflectivity gradient changes across the whole 3D volume. Image processing and modern analytical techniques such as dynamic Finite State Automata and Gaussian Process arecombined to identify and align ‘layers’ in volumes and perform feature detection. Reconstructed layers are analyzed to extract features such as vessels; location and shape of the optic cup; tilt of optic disc and retinal nerve fibre layer (RNFL). The analysis is conducted in a probabilistic framework without user input or a priori information. The algorithm was evaluated on images acquired on RTVue (Optovue, CA) using 4×4mm 3D disk scan from 26 glaucomatous and 14 normal subjects, with each eye scanned 3 times. RNFL thickness (RNFLT) was estimated at a wide annulus (0.58mm, 1.17mm; (width, inner margin radius)) and a narrow one (0.29mm, 1.32mm). Reproducibility (2×SD, and coefficient of variability (CV)), was calculated for quadrants.
3DSEGMENT segmented the retinal structures in 240 volume scans without spurious results. RNFLT maps are clinically realistic and highly reproducible (Table), yielding a significant improvement on statistics from StratusOCT imaging (e.g., Budenz et al 2005). Reproducibility was consistent regardless of annulus width (Table), suggesting that 3DSEGMENT is robust and stable across the 3D volume. Moreover, the quadrant RNFLT indicate separation between normal and glaucomatous subjects (Table).
FD-OCT provides volumetric images of retina with unprecedented detail. Quantitative analysis of these images is a significant challenge. Accurate segmentation algorithms and quantified features in these images are needed to realise the full potential of this technology: 3DSEGMENT serves this purpose and produces reliable estimates of RNFLT that can be used in the diagnosis and monitoring progression of glaucoma.
This PDF is available to Subscribers Only