Diagnostic accuracy of individual and combined retinal layer thickness measurements assessed at each pixel within posterior pole OCT scans of healthy and glaucomatous eyes

Abstract

Purpose : To compare the ability of individual and combined inner retinal layer thickness measurements to discriminate glaucomatous (GL) from healthy control (HC) eyes on a pixel-by-pixel basis.

Methods : GL eyes were included if 24-2 visual field (VF) pattern standard deviation or glaucoma hemifield test was outside normal limits on 2 consecutive tests. Posterior pole OCT scans (Spectralis, Heidelberg Engineering, GmbH) were obtained in 188 GL and 355 HC eyes (61 B-scans, 768 A-lines each, spanning total area of 30°×25° centered on the fovea and oriented parallel to the Fovea-Bruch’s Membrane Opening (FoBMO) axis. Segmentations within each B-scan were manually corrected and thickness values were measured for individual inner retinal layers: nerve fiber layer (NFL), ganglion cell layer (GCL), inner plexiform layer (IPL), as well as for the ganglion cell complex (GCC) and the total neural retina (TR). Values at each pixel were adjusted to the mean age and FoBMO distance of HC eyes. For each eye, the number of pixels with thickness values <5^th percentile of the HC distribution was used to create a receiver operating characteristic (ROC) curve for each layer and for layer combinations.

Results : The median VF mean deviation (MD) in GL was -4.37 dB (range: −27.0 to 0.72 dB). The individual inner retinal layers with the highest area under the ROC curve (AUROC) were the NFL and GCL; IPL performance was significantly lower (Table 1, Fig 1). GCC had a slightly (but not significantly) higher AUROC (94.4%) than either NFL or GCL (92.8% and 92.6%), but GCC accuracy was significantly higher than IPL (86.9%) and TR (90.5%). The highest AUROC (94.8%) was found for a combination criterion NFL or GCL, which had 83% sensitivity at 95% specificity.

Conclusions : Using pixel-based thickness measurements, the diagnostic accuracy of the NFL and GCL exceeded that of IPL and TR. GCC had marginally higher AUROC than NFL but lower sensitivity at 95% specificity. Boolean combinations of individual layers provided only a modest (and statistically insignificant) improvement in diagnostic accuracy. The best posterior pole OCT parameters were comparable to minimum rim width and peripapillary NFL in a similar cohort (Yang et al AJO in press). Combining anatomic sites in a topographically linked manner may further improve the diagnostic accuracy of OCT in glaucoma.

This is a 2020 ARVO Annual Meeting abstract.