Impact of segmentation density on Spectral Domain Optical Coherence Tomography (SD-OCT) assessments in Geographic Atrophy

Abstract

Purpose : To assess the influence of varying B-scan frame-sampling densities on outer retinal layers thickness and area measurements from spectral domain optical coherence tomography (OCT) in eyes with geographic atrophy (GA) secondary to dry age related macular degeneration.

Methods : 30 eyes from 30 subjects with GA were recruited from Doheny-UCLA Eye Centers and enrolled in this IRB-approved retrospective study. Spectralis SD-OCT volume scans (512 X 49 macular cube over 20X20 degrees) were obtained for all subjects. All 49 B-scans were segmented manually using reading center software (3D-OCTOR) to quantify total retina, outer nuclear layer (ONL), photoreceptor inner segments, photoreceptor outer segments, and the retinal pigment epithelial layer (RPE). Mean thickness (MT) and total area (TA) were generated using all 49 B-scans (spaced 122 µm apart), 25 B-scans (every other B-scan, spaced 240 µm apart), and 17 B-scans (every third scan, 353 µm apart). Values generated using reduced scan densities were compared with the thickness and area measurements using all 49 B-scans (paired Sample t test; SPSS 18.0).

Results : For outer retinal layer thickness and area (except RPE area) measurements generated using all 49 B-scans were statistically similar (p>0.05) when compared to the values from reducing the scan density to every other B-scan (25 B-scans). However, when the scan density was further reduced to every third B-scan (17 B-scans), the differences in values compared to all 49 B-scans were significantly different (for inner segments and RPE; Table 1).

Conclusions : These data suggest that a selective segmentation strategy may be a viable alternative to exhaustive and comprehensive manual segmentation in geographic atrophy. Exhaustive and comprehensive manual segmentation may not be needed since using every other B-scan may yield results similar to ground truth (49 B-scans), except for RPE area.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.