Purchase this article with an account.
Jason L Chien, Mark P Ghassibi, Celso Tello, Jeffrey M Liebmann, Robert Ritch, Travis Porco, Robert L Stamper, Sung Chul (Sean) Park; Optimizing Macular Layer Grids in Spectral-Domain Optical Coherence Tomography for Glaucoma Diagnosis. Invest. Ophthalmol. Vis. Sci. 2016;57(12):870. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To find the macular grids with the best glaucoma diagnostic capability for ganglion cell layer (GCL), ganglion cell complex (GCC; retinal nerve fiber layer [RNFL]+GCL+inner plexiform layer), and retinal nerve fiber ganglion cell layer (RNFGCL; RNFL+GCL) in spectral-domain (SD) optical coherence tomography (OCT).
Serial horizontal SD OCT scans of the macula (30x25 degree rectangle; interval between scans, ~120 µm) were obtained using SD OCT in one eye of normal and glaucoma subjects. An automated posterior pole grid (64 squares; 8x8) centered on the fovea was used (Fig 1A). For the 64 squares, a penalized logistic regression model using the elastic net algorithm was performed for GCL, GCC, and RNFGCL to identify combinations of squares with the greatest area under the receiver operating characteristic curves (AUC) for glaucoma diagnosis. Squares in the regions previously identified with the greatest AUCs (Chien JL, et al. IOVS 2015;56:4552; Fig 1B-G) were constrained to be included. Absolute value (Least Absolute Shrinkage and Selection Operator method) and quadratic (Ridge method) penalties were equally weighted. Cross validation was used to identify the penalty weights producing the best predictive value. AUCs of identified square combinations were compared among each other. Analysis was conducted using GLM net package of R program (Version 3.1 for Mac R foundation for statistical computing, Vienna, Austria).
The AUCs of 6 identified combinations of squares ranged from 0.957 to 0.977 for GCL, 0.966 to 0.999 for GCC, and 0.968 to 0.994 for RNFGCL (AUCs were obtained with an exploratory model and the clinical application of these values awaits further validation on independent data). The macular grids with the best and the second best AUCs for GCL, GCC, and RNFGCL are shown in Figure 2. Among the 6 regions constrained to be included (Fig 1B-G), the combination of squares including both the temporal and inferior parafoveal regions (Fig 1F and 1G) generally yielded the highest AUCs (Fig 2A-D and 2F). The best AUCs of GCC and RNFGCL were significantly greater than that of GCL (p=0.023 and 0.032, respectively).
For GCL, GCC, and RNFGCL, macular grids should include the temporal and inferior parafoveal regions for better glaucoma diagnostic capability. GCC and RNFGCL appear to be more promising than GCL for glaucoma diagnosis.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
This PDF is available to Subscribers Only