Purchase this article with an account.
Julia Schottenhamml, Eric M. Moult, Eduardo Amorim Novais, Martin F Kraus, ByungKun Lee, WooJhon Choi, Stefan B Ploner, Lennart Husvogt, Chen D Lu, Patrick Yiu, Philip J Rosenfeld, Jay S Duker, Andreas K Maier, Nadia Waheed, James G Fujimoto; OCT-OCTA Segmentation: a Novel Framework and an Application to Segment Bruch's Membrane in the Presence of Drusen. Invest. Ophthalmol. Vis. Sci. 2017;58(8):645.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We present a novel framework for segmenting optical coherence tomography (OCT) and OCT angiography (OCTA) that jointly uses structural and angiographic information. We term this new paradigm “OCT-OCTA segmentation,” and demonstrate its utility by segmenting Bruch’s membrane (BM) in the presence of drusen.
We developed an automatic OCT-OCTA graph-cut algorithm for BM segmentation. Our algorithm’s performance was quantitatively validated by comparing it with manual segmentation in 7 eyes (6 patients; 73.8±5.7 y/o) with drusen. The algorithm was also qualitatively assessed in healthy eyes (n=13), eyes with diabetic retinopathy (n=21), early/intermediate age-related macular degeneration (AMD) (n=14), exudative AMD (n=5), geographic atrophy (GA) (n=6), and polypoidal choroidal vasculopathy (n=7).
The absolute pixel-wise error between the manual and automatic segmentations had the following values: mean: 4.5±0.89um; 1st Quartile: 1.9±1.35um; 2nd Quartile: 3.9±1.90um; and 3rd Quartile: 6.3±2.67. This corresponds to a mean absolute error smaller than the optical axial resolution of our OCT system (~8-9um). In all other tested eyes, qualitative visual inspection showed BM contours that were deemed suitably accurate for use in forming en face OCT(A) projections. The algorithm’s poorest results occurred in GA patients with large areas of atrophy.
By leveraging both structural and angiographic information we showed that OCT-OCTA segmentation is likely to be a widely useful framework for segmenting ocular structures.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.
En face segmentation analysis, where each column corresponds to a different nGA/DAGA eye. 1st row: color fundus photo; 2nd, 3rd rows: en face OCTA slice taken at the manual and automatic segmentations, respectively; 4th, 5th row: en face OCT slices taken the manual and automatic segmentations, respectively. 6th row: heat map of the segmentation error (legend, bottom right; units of pixels; 1 pixel = 4.5um). All OCT(A) fields are 6x6mm.
OCT(A) B-scan analysis of manual (teal) and automatic (orange) segmentations; A-D are taken from the white lines in Figure 1. Each row shows both OCT data (left) and OCTA data (right). Enlargements, indicated by red and green boxes, are also shown.
This PDF is available to Subscribers Only