The ability of deep learning architectures to identify glaucomatous optic neuropathy (GON) in fundus photographs was evaluated. A large database of fundus photographs (n = 14,822) from a racially and ethnically diverse group of individuals (over 33% of African descent) was evaluated by expert reviewers and classified as GON or healthy. Several deep learning architectures and the impact of transfer learning were evaluated. The best performing model achieved an overall area under receiver operating characteristic (AUC) of 0.91 in distinguishing GON eyes from healthy eyes. It also achieved an AUC of 0.97 for identifying GON eyes with moderate-to-severe functional loss and 0.89 for GON eyes with mild functional loss. A sensitivity of 88% at a set 95% specificity was achieved in detecting moderate-to-severe GON. In all cases, transfer improved performance and reduced training time. Model visualizations indicate that these deep learning models relied on, in part, anatomical features in the inferior and superior regions of the optic disc, areas commonly used by clinicians to diagnose GON. The results suggest that deep learning-based assessment of fundus images could be useful in clinical decision support systems and in the automation of large-scale glaucoma detection and screening programs.

Objective Conventional optic disc margin-based neuroretinal rim measurements lack a solid anatomical and geometrical basis. An optical coherence tomography (OCT) index, Bruch’s membrane opening minimum rim width (BMO-MRW), addresses these deficiencies and has higher diagnostic accuracy for glaucoma. We characterized BMO-MRW and peripapillary retinal nerve fiber layer thickness (RNFLT) in a normal population. Design Multi-centred cross-sectional study. Participants Normal White subjects. Methods Approximately equal number of subjects in each decade group (20–90 years) was enrolled in 5 centers. Subjects had normal ocular and visual field examinations. We obtained OCT images of the optic nerve head (24 radial scans) and peripapillary retina (1 circular scan). The angle between the fovea and BMO center (FoBMO), relative to the horizontal axis of the image frame, was first determined and all scans were acquired and analyzed relative to this eye-specific FoBMO axis. Variation of BMO-MRW and RNFLT was analyzed with respect to age, sector and BMO shape. Main Outcome Measures Age-related decline and between-subject variability in BMO-MRW and RNFLT. Results There were 246 eyes of 246 subjects with a median age of 52.9 (range, 19.8 to 87.3) years. The median FoBMO angle was −6.7° (range, 2.5° to −17.5°). BMO was predominantly vertically oval with a median area of 1.74 mm2 (range, 1.05 to 3.40 mm2). Neither FoBMO angle nor BMO area was associated with age or axial length. Both global mean BMO-MRW and RNFLT declined with age at a rate of −1.34 µm/y and −0.21 µm/y, equivalent to 4.0% and 2.1% loss per decade of life, respectively. Sectorally, the most rapid decrease occurred inferiorly and the least temporally, however, the age association was always stronger with BMO-MRW than with RNFLT. There was a modest relationship between mean global BMO-MRW and RNFLT (r = 0.35), while sectorally the relationship ranged from moderate (r = 0.45, inferotemporal) to non-existent (r = 0.01, temporal). Conclusions There was significant age-related loss of BMO-MRW in healthy subjects and notable differences between BMO-MRW and RNFLT in their relationship with age and between each other. Adjusting BMO-MRW and RNFLT for age and sector is important in ensuring optimal diagnostics for glaucoma.

Purpose To compare the rates of retinal nerve fiber layer (RNFL) loss in patients suspect of having glaucoma who developed visual field damage (VFD) to those who did not develop VFD, and to determine whether the rate of RNFL loss can be used to predict who will develop VFD.. Design Prospective observational cohort study Participants Glaucoma suspects, defined as having glaucomatous optic neuropathy or ocular hypertension (Intraocular pressure (IOP)>21 mmHg) without repeatable VFD at baseline from the Diagnostic Innovations in Glaucoma Study, and the African Descent and Glaucoma Evaluation Study. Methods Global and quadrant RNFL thickness (RNFLT) were measured with the Spectralis spectral-domain optical coherence tomography (SD-OCT). VFD was defined as having 3 consecutive abnormal visual fields. The rate of RNFL loss in eyes developing VFD was compared with eyes not developing VFD using multivariable linear mixed-effects models. A joint longitudinal survival model utilized the estimated RNFLT slope to predict the risk of developing VFD, while adjusting for potential confounding variables. Main Outcome Measures The rate of RNFL thinning and the probability of developing VFD. Results Four hundred and fifty-four eyes of 294 glaucoma suspects were included. The average number of SD-OCT examinations was 4.6 (range, 2–9) with median follow-up time of 2.2 (0.4–4.1) years. Forty eyes (8.8%) developed VFD. The estimated mean rate of global RNFL loss was significantly faster in eyes developing VFD compared with eyes that did not (−2.02μm/year vs. −0.82μm/year, P<0.001). The joint longitudinal survival model showed that each 1μm/year faster rate of global RNFL loss corresponded to a 2.05 times higher risk of developing VFD (Hazards Ratio (HR)=2.05, 95% Confidence Interval (CI): 1.14–3.71; p=0.017). Conclusions The rate of global RNFL loss was more than twice as fast in eyes developing VFD compared with eyes that did not develop them. Joint longitudinal survival model showed that a 1μm/year faster rate of RNFLT loss corresponded to a 2.05 times higher risk of developing VFD. These results suggest that measuring the rate of SD-OCT RNFL loss may be a useful tool to help identify patients who are at a high risk of developing visual field loss.

Purpose To investigate the prevalence of visual field defects in glaucomatous eyes, glaucoma suspects, and ocular hypertensives with 24-2 and 10-2 visual fields. Design Prospective, cross-sectional study. Participants Patients with or suspected glaucoma. Testing 24-2 and 10-2 visual fields. Patients were classified into 3 groups based upon the presence of glaucomatous optic neuropathy (GON) and 24-2 visual field abnormalities: early glaucoma (GON and abnormal visual field, mean deviation >−6 dB), glaucoma suspects (GON and normal visual field), and ocular hypertensives (normal disc, normal visual field, and intraocular pressure >22 mmHg). For the classification of visual field abnormalities, 24-2 and 10-2 tests performed on the same visit were analyzed. Main outcome measure Comparison of the prevalence of abnormal 24-2 versus 10-2 visual field results based upon cluster criteria in each diagnostic group. Results 775 eyes (497 patients) were evaluated. 364 eyes had early glaucoma, 303 were glaucoma suspects, and 108 were ocular hypertensives. In the glaucoma group, 16 of the 26 (61.5%) eyes classified as normal based upon cluster criteria on 24-2 tests were classified as abnormal on 10-2 visual fields. In eyes with suspected glaucoma, 79 of the 200 (39.5%) eyes classified as normal on 24-2 were classified as abnormal on 10-2 visual fields. In ocular hypertensive eyes, 28 of the 79 (35.4%) eyes classified as normal on the 24-2 were abnormal on the 10-2. Patients of African descent were more likely to have an abnormal 10-2 result (67.3 vs. 56.8%, P=0.009). Conclusions Central visual field damage seen on 10-2 is often missed with 24-2 strategy in all groups. This finding has implications for the diagnosis of glaucoma and classification of severity.

No abstract