Detection of Open Angle Glaucoma with Patient-Specific Macular Structural Measurements

Abstract

Purpose : Glaucomatous damage is most commonly interpreted using ONH based markers, however, growing evidence have shown the role of macular indices in the assessment of glaucoma. In this study, we compare new cubic grid with the standard circular grid scan pattern, as well as test our new Bayesian method in identifying patient-specific structural measures in differentiating between healthy and glaucoma eyes.

Methods : Fifty-six healthy eyes and 73 glaucoma eyes (mean visual field mean deviation= -3.7dB) were included. Spectralis OCT cubic grid macular scans were analyzed in 2 ways; as a 4mm x 4mm with an excluded center of 1.5mm x 1.5mm, and as the standard circular grid scans with a diameter of 3.45mm. The minimum ganglion cell layer (GCL) and the ganglion cell-inner plexiform layer (GCIPL) thicknesses of the 4 quadrants of the cube grid takes into account the subject’s defect location, the minimum thickness measurement that corresponds to the edge of the structural defect. In addition, Bayesian methods were used to calculate the probability of each structural parameter for discriminating between healthy and OAG eyes. Variables with higher probabilities were selected as the best input for the classifier which allowed us to select individual-based defect locations and structural measurements with the best discrimination ability. Area under the receiver operating characteristic (AUROC) curves were used to compare the diagnostic accuracy of Bayesian selection methods and different macular thickness measurements.

Results : The cubic grid (Figure 1A) when compared to the standard circular grid showed improved discrimination between healthy and OAG eyes for both GCIPL and GCL thickness measurements. The AUROC of the 1) minimum thickness measures had similar diagnostic accuracy to 2) global mean values for both GCL (1) AUROC: 0.92 and 2) 0.91, respectively (p-value=0.08) and GCIPL (1) AUROC: 0.92 and 2) 0.91, respectively (p-value=0.10). However, using the Bayesian selection of the best measurements, we obtained significantly better discrimination (AUROC=0.94) than mean GCL or GCIPL alone (p-value=0.02) (Figure 1B).

Conclusions : The new macular cubic grid scan facilitates the identification of patient-specific structural measures and can improve the diagnostic power of OAG detection compared to the standard methods.

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.