Purchase this article with an account.
Sabrina Khan Khalil, Alexander Shahin, David Richards, Ramesh Ayyala, Radouil T Tzekov; Variability in outer retina layers segmentation by SD-OCT. Invest. Ophthalmol. Vis. Sci. 2020;61(7):2576.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To investigate variability and replicability of outer retinal macular layers’ segmentation by two models of Spectralis SD-OCT (Heidelberg Engineering Inc., Carlsbad, CA).
Posterior pole SD-OCT scans (8 x 8 grid) centered on the fovea were performed on normal volunteers. Three consecutive images of each eye were acquired on two models (2011 and 2017) of Spectralis SD-OCT by two operators. The same procedure was repeated in 2-3 weeks. The inner nuclear layer (INL), the outer plexiform layer (OPL), the outer nuclear (ONL) and the retinal pigment epithelial layer (RPE) were segmented automatically by Spectralis software. For each grid cell, the root mean square (RMS) deviation from the mean of each set of three measurements, for each layer, was calculated and analyzed both in terms of absolute values and as a percent of the average thickness at that location.
Both eyes of 14 volunteers (4 men and 10 women) aged 35.1 +/- 12.3 years were imaged. Without any algorithm correction, the RMS deviation among the three images varied as follows: 0.6-1.8 µm (INL), 0.6-2.4 µm (OPL), 0.5-2.8 µm (ONL), 0.3-1.6 µm (RPE) and there was no difference between right and left eyes (p>0.05). These ranges translated to 2.1-10.1% (INL), 2.1-7.5% (OPL), 0.7-5.9% (ONL) and 1.9-5.2% (RPE) of local retinal thickness. Algorithm failure, defined as variability of more than 10% of RMS error in retinal thickness between the 3 measurements, occurred in 5.4% of the cells (INL), 7.5% (OPL), 2.9% (ONL) and 3.0% (RPE), and in 15 out of 16 comparisons there was no difference between right and left eyes (p>0.05). When algorithm failure was accounted for, the average RMS error of local retinal thickness was 3.2% (INL), 3.3% (OPL), 1.8% (ONL) and 3.0% (RPE) with no difference between right and left eyes (p>0.05), while the average thickness of the layers was 31.9 ± 1.6 µm (INL), 25.9 ± 1.5 µm (OPL), 56.7 ± 5.2 µm (ONL) and 13.1 ± 1.0 µm (RPE) with no difference between right and left eyes (p>0.05).
These findings indicate that algorithm failure in segmentation of outer retina layers in the macula is relatively infrequent (< 8%) in healthy volunteers with some variability between different layers. In areas where the algorithm performed well, the average RMS error was relatively low (< 3.4%).
This is a 2020 ARVO Annual Meeting abstract.
This PDF is available to Subscribers Only