Editors Selection IGR 20-2

Clinical Examination Methods: Macular Findings in Early Glaucoma

Gustavo de Moraes
Bruna Melchior Silva

Comment by Gustavo de Moraes & Bruna Melchior Silva on:

79863 A Deep Learning Algorithm to Quantify Neuroretinal Rim Loss From Optic Disc Photographs, Thompson AC; Jammal AA; Medeiros FA, American Journal of Ophthalmology, 2019; 201: 9-18

Find related abstracts

Fundus photography is one of the most common methods to assess the presence of glaucomatous optic neuropathy, although its interpretation by human graders is highly subjective and studies have shown modest reproducibility and interrater reliability.1 However, the interest in fundus photographs has increased due to recent studies that have demonstrated the ability of deep learning algorithms (DLA) to provide accurate analysis of retinal diseases based on them. 2 Moreover, fundus photographs are relatively inexpensive and largely available worldwide, which make their use particularly promising for community screening.

In this study, Thompson et al. developed a novel DLA that has been trained on using the Bruch's membrane opening minimum rim width (BMO-MRW) from spectral domain (SD) OCT to detect glaucomatous optic neuropathy on fundus photographs and to quantitatively predict the amount of neuroretinal tissue. Studies have suggested that BMO-MRW is an accurate parameter for differentiating glaucomatous from healthy eyes and has good correlation with visual field (VF) loss. 3 Moreover, this objective parameter can minimize human mistakes that frequently happen when subjectively assessing the limits of the optic nerve and estimating the amount of neural tissue that exits the globe.

The study included 9,282 pairs of optic disc photographs and SDOCT scans from 927 eyes of 490 subjects, divided into normal, suspect and glaucomatous. In the independent test set, the mean DLA BMO-MRW predictions of global rim area had a strong correlation with the observed values of actual SDOCT BMO-MRW (r=0.88, p<0.001). They also evaluated the relationship between the predicted values of sectorial BMO-MRW and the corresponding sectoral VF sensitivities according to the Garway-Heath structure-function map, which had similar strength to those observed for the actual SDOCT BMO-MRW measurements. The global BMO-MRW predictions were also significantly associated with VF MD (r=0.49, p<0.01). This is an important study for the objective quantification of neuroretinal damage from fundus photographs, which could potentially reduce the rates of undiagnosed glaucoma worldwide. Moreover, by overcoming issues related with the reproducibility of subjective disc photo evaluation, it could be useful to improve estimates of structural progression based on photos.


  1. Varma R, Steinmann WC, Scott IU. Expert agreement in evaluating the optic disc for glaucoma. Ophthalmology 1992; 99(2):215-221.
  2. Son J et al. Development and Validation of Deep Learning Models for Screening Multiple Abnormal Findings in Retinal Fundus Images. Ophthalmology. 2019 May 31. pii: S0161-6420(19)30374-4.
  3. Danthurebandara VM, Vianna JR, Sharpe GP, Hutchison DM, Belliveau AC, Shuba LM, Nicolela MT, Chauhan BC. Diagnostic Accuracy of Glaucoma With Sector-Based and a New Total Profile-Based Analysis of Neuroretinal Rim and Retinal Nerve Fiber Layer Thickness. Invest Ophthalmol Vis Sci. 2016;57(1):181-7

Issue 20-2

Select Issue