Neurovascular segregation of the retinal nerve fiber layer in glaucoma

The imaging data of one eye from 154 healthy and 143 glaucoma participants were acquired to evaluate the contributions of the neuronal and vascular components within the retinal nerve fiber layer (RNFL) for detecting glaucoma and modeling visual field loss through the use of optical coherence tomography (OCT) and OCT angiography. The neuronal and vascular components within the circumpapillary RNFL were independently evaluated. In healthy eyes, the neuronal component showed a stronger association with age (r = -0.52, p < 0.001) compared to measured RNFL thickness (r = -0.46, p < 0.001). Using the neuronal component alone improved detection of glaucoma (AUC: 0.890 ± 0.020) compared to measured RNFL thickness (AUC: 0.877 ± 0.021; χ2 = 5.54, p = 0.019). Inclusion of the capillary components with the sectoral neuronal component resulted in a significant improvement in glaucoma detection (AUC: 0.927 ± 0.015; χ2 = 15.34, p < 0.001). After adjusting for potential confounders, AUC increased to 0.952 ± 0.011. Results from modeling visual field loss in glaucoma eyes suggest that visual field losses associated with neuronal thinning were moderated in eyes with a larger capillary component. These findings suggest that segregation of the neurovascular components could help improve understanding of disease pathophysiology and affect disease management in glaucoma.

The role of the disc damage likelihood scale in glaucoma detection by community optometrists

PURPOSE

Recent guidelines recommend disc damage likelihood scale (DDLS) is recorded for all referrals of suspected glaucoma from community optometrists to hospital eye services (HES) in Scotland. This study aimed to determine whether lower DDLS grades were associated with higher rates of discharge at the first visit to HES.

METHODS

A retrospective analysis of 618 consecutive new referrals from community optometrists to a university hospital glaucoma service. 65 (10.5%) included DDLS graded by the community optometrist. A comprehensive eye examination and optical coherence tomography (OCT) was performed in the hospital glaucoma clinic and first visit discharge rate (FVDR) for different grades of DDLS compared. The relationship between DDLS and retinal nerve fibre layer (RNFL) thickness on OCT was also examined.

RESULTS

The FVDR for patients with DDLS recorded in the referral was 27.7% (18 of 65) compared to 25% (138 of 553) in those without DDLS (p = 0.631). The FVDR was 50% for those with a DDLS of 3 in the worse eye, decreasing to 32% and 21% for DDLS grades of 4 and 5 respectively. No patient with a DDLS ≥ 6 was discharged at the first visit and none with a DDLS < 4 (the cut off for consideration of referral in Scottish guidelines) were found to have glaucoma. There was a significant but weak inverse relationship between DDLS and RNFL thickness. The strongest relationship was with average RNFL thickness (r = -0.378, p < 0.01) followed by superotemporal (r = -0.359, p < 0.01) and inferotemporal (r = -0.353, p < 0.01) RNFL thickness.

CONCLUSIONS

In patients referred to HES with DDLS information included, lower DDLS grading was associated with higher odds of being discharged at the first visit. DDLS grading by community optometrists had a poor correlation with RNFL thickness measured using OCT.

Accurate prediction of glaucoma from colour fundus images with a convolutional neural network that relies on active and transfer learning

PURPOSE

To assess the use of deep learning (DL) for computer-assisted glaucoma identification, and the impact of training using images selected by an active learning strategy, which minimizes labelling cost. Additionally, this study focuses on the explainability of the glaucoma classifier.

METHODS

This original investigation pooled 8433 retrospectively collected and anonymized colour optic disc-centred fundus images, in order to develop a deep learning-based classifier for glaucoma diagnosis. The labels of the various deep learning models were compared with the clinical assessment by glaucoma experts. Data were analysed between March and October 2018. Sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and amount of data used for discriminating between glaucomatous and non-glaucomatous fundus images, on both image and patient level.

RESULTS

Trained using 2072 colour fundus images, representing 42% of the original training data, the trained DL model achieved an AUC of 0.995, sensitivity and specificity of, respectively, 98.0% (CI 95.5%-99.4%) and 91% (CI 84.0%-96.0%), for glaucoma versus non-glaucoma patient referral.

CONCLUSIONS

These results demonstrate the benefits of deep learning for automated glaucoma detection based on optic disc-centred fundus images. The combined use of transfer and active learning in the medical community can optimize performance of DL models, while minimizing the labelling cost of domain-specific mavens. Glaucoma experts are able to make use of heat maps generated by the deep learning classifier to assess its decision, which seems to be related to inferior and superior neuroretinal rim (within ONH), and RNFL in superotemporal and inferotemporal zones (outside ONH).

Evaluation of two-dimensional Bruch's membrane opening minimum rim area for glaucoma diagnostics in a large patient cohort

PURPOSE

To characterize the two-dimensional parameter Bruch's membrane opening minimum rim area (BMO-MRA) in spectral domain optical coherence tomography (SD-OCT) of the optic nerve head (ONH) compared to minimum rim width (BMO-MRW) and retinal nerve fibre layer (RNFL) thickness in a large patient cohort.

METHODS

Case-control, cross-sectional study of 705 eyes of 445 participants. A total of 449 eyes with glaucoma, 67 eyes with ocular hypertension and 189 healthy controls, underwent SD-OCT and confocal laser scanning tomography (CSLT), visual field testing and clinical examination. Morphometric ONH parameters, visual field function and diagnostic power were compared. Main outcome measures were SD-OCT-derived BMO-MRA, BMO-MRW, RNFL thickness and CSLT-derived rim area (DM-RA).

RESULTS

Mean ONH area was 2.11 ± 0.57 mm² ; mean BMO area was 1.89 ± 0.45 mm² . Correlation of mean deviation in visual field to morphometric parameters was ρ = 0.70, (p < 0.001) for RNFL thickness, ρ = 0.68 (p < 0.001) for BMO-MRA, ρ = 0.66 (p < 0.001) for BMO-MRW. These correlations were not significantly different (p > 0.05), while DM-RA correlated significantly worse (ρ = 0.55; p < 0.001). In receiver operating characteristics, the calculated area under the curve (AUC) and sensitivity at 90% specificity to differentiate glaucoma were 0.87% and 70.1% for BMO-MRA, 0.86% and 68.1% for RNFL thickness, 0.84% and 66.0% for BMO-MRW, 0.82% and 51.3% for DM-RA.

CONCLUSIONS

In a heterogenous clinical cohort of glaucoma patients, all analysed SD-OCT parameters excel DM-RA of CSLT. The two-dimensional parameter BMO-MRA shows comparable levels of diagnostic power to detect glaucoma compared to established parameters BMO-MRW and RNFL thickness. Given higher comparability between ONH sizes, BMO-MRA might become an additional standard tool in SD-OCT imaging for glaucoma.

Improving Access to Eye Care among Persons at High-Risk of Glaucoma in Philadelphia--Design and Methodology: The Philadelphia Glaucoma Detection and Treatment Project

PURPOSE

The Wills Eye Glaucoma Research Center initiated a 2-year demonstration project to develop and implement a community-based intervention to improve detection and management of glaucoma in Philadelphia.

METHODS

The glaucoma detection examination consisted of: ocular, medical, and family history; visual acuity testing; corneal pachymetry; biomicroscopy of the anterior segment; intraocular pressure (IOP) measurement; gonioscopy; funduscopy; automated visual field testing; and fundus-color photography. Treatment included laser surgery and/or IOP-lowering medication. A cost analysis was conducted to understand resource requirements. Outcome measures included; prevalence of glaucoma-related pathology and other eye diseases among high-risk populations; the impact of educational workshops on level of knowledge about glaucoma (assessed by pre- and post-test evaluation); and patient satisfaction of the glaucoma detection examinations in the community (assessed by satisfaction survey). Treatment outcome measures were change in IOP at 4-6 weeks and 4-6 months following selective laser trabeculoplasty treatment, deepening of the anterior chamber angle following laser-peripheral iridotomy treatment, and rate of adherence to recommended follow-up examinations. Cost outcomes included total program costs, cost per case of glaucoma detected, and cost per case of ocular disease detected.

RESULTS

This project enrolled 1649 participants (African Americans aged 50+ years, adults 60+ years and individuals with a family history of glaucoma). A total of 1074 individuals attended a glaucoma educational workshop and 1508 scheduled glaucoma detection examination appointments in the community setting.

CONCLUSIONS

The Philadelphia Glaucoma Detection and Treatment Project aimed to improve access and use of eye care and to provide a model for a targeted community-based glaucoma program.

Disease severity in newly diagnosed glaucoma patients with visual field loss: trends from more than a decade of data

PURPOSE

Large archives of visual field (VF) records from automated perimetry are used to examine severity of vision loss at diagnosis in glaucoma patients over a 13 year period in England.

METHODS

A total of 473 252 Humphrey VFs recorded at four regionally different glaucoma clinics in England were retrospectively examined. Patients were required to have a Humphrey mean deviation (MD) outside 95% normative limits in at least one eye and had at least two visits to the glaucoma clinic in a study period between the start of 1999 and the end of 2011. No other clinical data was considered. MD of the worse eye at the first hospital visit was used as an estimate of vision loss for a patient at diagnosis; proportion of patients categorised as having 'early' (better than -6 dB) or 'advanced' (worse than -12 dB) VF loss were also calculated. Simple linear regression of MD against the date of first visit was used to estimate changes in vision loss at diagnosis over time.

RESULTS

Median age, at the time of diagnosis, of the 25 521 patients included in the analysis was 67 (interquartile range 55-76) years. Average level of glaucoma vision loss at diagnosis, in those patients presenting to secondary care with a VF defect, improved by an average 0.11 dB per year over the study period (95% confidence interval: 0.08-0.13 dB per year; p < 0.0001). Percentage of patients with 'advanced' VF loss in at least one eye at diagnosis changed from 30% (1999-2001) to 21% in (2009-2011) (p < 0.0001).

CONCLUSIONS

Severity of vision loss at the point of glaucoma detection, in those patients diagnosed with a VF defect, is improving over time in England. Nevertheless, the improvement is modest and large numbers of patients still present at glaucoma clinics with significant vision loss in at least one eye. Large scale digital VF data can be used to help monitor and audit health service delivery of glaucoma.

False negative results in glaucoma detection with Heidelberg Retina Tomograph II

Purpose

To evaluate the rate of false negative results with the Heidelberg Retina Tomograph (HRT II) in a glaucoma practice.

Design

Cross-sectional study.

Methods

We analyzed the HRTs taken between October 2002 and October 2003 in our glaucoma clinic, and selected the patients who had a good quality image (SD < 40 μ) with a normal Moorfield’s Regression Analysis (MRA). A masked independent observer classified those patients as normal, glaucoma suspect, or glaucomatous on the basis of optic disc stereo photos (ODP) and at least 2 consecutive reliable automated perimetries. The diagnosis of glaucoma was based on a glaucomatous optic disc with a congruent, reproducible visual field defect.

Results

Four hundred and fifty patients who had undergone an HRT examination were analyzed. One hundred and nine patients had an HRT classified as normal on the MRA, and a good quality image. Fifteen of those 109 patients (13.7%) were classified as glaucomatous on the basis of an abnormal ODP with corresponding visual field defect. Seven (6.4%) patients were classified as glaucoma suspect.

Conclusion

Fourteen percent of glaucoma patients with glaucoma remained undetected with the HRT II Moorfield’s regression analysis as a sole means to detect glaucoma.