Abstract
Purpose :
OCT provides the ability to measure the thickness of the ganglion cell plus inner plexiform layers (GCIPL) as well as visualizing the various tissues within the retina. GCIPL thickness is measured as the difference between the outer boundary of the retinal nerve fiber layer (RNFL) and the outer boundary of the inner plexiform layer (IPL). We have developed a multi-retinal layer segmentation algorithm (MLS) that segments the RNFL and IPL as well as other layers. The purpose of this study is to compare the performance of the GCIPL analysis using MLS and a commercial algorithm.
Methods :
Subjects with no retinal disease and with glaucoma were scanned with the 200x200 and 512x218 macular cube scans over 6x6 mm using CIRRUS™ HD-OCT 5000 (ZEISS, Dublin, CA). Three B-scans (at 2.2 mm, 3 mm, 3.9 mm) of 70 512x128 scans were used to evaluate the accuracy of RNFL and IPL using MLS and CIRRUS algorithm. Bland-Altman analysis was performed to compare the two automated segmentations to manual segmentation. GCIPL thickness maps were generated. All GCIPL maps of OD eyes were flipped along the y-axis to match the GCIPL maps of OS eyes. The area under the curve (AUC) was evaluated using a combined dataset that included 396 normal subjects and 259 glaucoma subjects. For each of these subjects, at least 3, and as many as 9, repeated scans were acquired, and from these, repeatability and coefficient of variance were calculated.
Results :
The AUC was calculated in all subfields, where MLS shows greater AUC in all sectors. The repeatability of the GCIPL layer thicknesses of normal and glaucoma cases was calculated separately for all subfields (Figure 1). MLS shows better repeatability in all sectors for normal cases and equal or better repeatability in four sectors for glaucoma cases. The overall mean difference between MLS and manual segmentations for RNFL and IPL was found to be less than ± 1.2 µm, while for the CIRRUS error case it was ± 2.5 µm (Figure 2). The limits of agreement with manual segmentation (± 1.96 times standard deviation) are also reported.
Conclusions :
This proposed MLS algorithm can achieve smaller mean difference and limits in both RNFL and IPL segmentation. Furthermore, MLS shows a better repeatability in almost all sectors for normal and glaucoma cases. It is therefore a promising tool to further improve existing OCT imaging devices.
This is a 2020 ARVO Annual Meeting abstract.