Prediction of Visual Field Progression with Baseline and Longitudinal Structural Measurements Using Deep Learning.

PURPOSE: Identifying glaucoma patients at high risk of progression based on widely available structural data is an unmet task in clinical practice. We test the hypothesis that baseline or serial structural measures can predict visual field (VF) progression with deep learning (DL). DESIGN: Development of a DL algorithm to predict VF progression. METHODS: 3,079 eyes (1,765 patients) with various types of glaucoma and ≥5 VFs, and ≥3 years of follow-up from a tertiary academic center were included. Serial VF mean deviation (MD) rates of change were estimated with linear-regression. VF progression was defined as negative MD slope with p<0.05. A Siamese Neural Network with ResNet-152 backbone pre-trained on ImageNet was designed to predict VF progression using serial optic-disc photographs (ODP), and baseline retinal nerve fiber layer (RNFL) thickness. We tested the model on a separate dataset (427 eyes) with RNFL data from different OCT. The Main Outcome Measure was Area under ROC curve (AUC). RESULTS: Baseline average (SD) MD was 3.4 (4.9)dB. VF progression was detected in 900 eyes (29%). AUC (95% CI) for model incorporating baseline ODP and RNFL thickness was 0.813 (0.757-0.869). After adding the second and third ODPs, AUC increased to 0.860 and 0.894, respectively (p<0.027). This model also had highest AUC (0.911) for predicting fast progression (MD rate <1.0 dB/year). Model's performance was similar when applied to second dataset using RNFL data from another OCT device (AUC=0.893; 0.837-0.948). CONCLUSIONS: DL model predicted VF progression with clinically relevant accuracy using baseline RNFL thickness and serial ODPs and can be implemented as a clinical tool after further validation.

Efficacy of Smoothing Algorithms to Enhance Detection of Visual Field Progression in Glaucoma.

Purpose: To evaluate and compare the effectiveness of nearest neighbor (NN)- and variational autoencoder (VAE)-smoothing algorithms to reduce variability and enhance the performance of glaucoma visual field (VF) progression models. Design: Longitudinal cohort study. Subjects: 7150 eyes (4232 patients), with ≥ 5 years of follow-up and ≥ 6 visits. Methods: Vsual field thresholds were smoothed with the NN and VAE algorithms. The mean total deviation (mTD) and VF index rates, pointwise linear regression (PLR), permutation of PLR (PoPLR), and the glaucoma rate index were applied to the unsmoothed and smoothed data. Main Outcome Measures: The proportion of progressing eyes and the conversion to progression were compared between the smoothed and unsmoothed data. A simulation series of noiseless VFs with various patterns of glaucoma damage was used to evaluate the specificity of the smoothing models. Results: The mean values of age and follow-up time were 62.8 (standard deviation: 12.6) years and 10.4 (standard deviation: 4.7) years, respectively. The proportion of progression was significantly higher for the NN and VAE smoothed data compared with the unsmoothed data. VF progression occurred significantly earlier with both smoothed data compared with unsmoothed data based on mTD rates, PLR, and PoPLR methods. The ability to detect the progressing eyes was similar for the unsmoothed and smoothed data in the simulation data. Conclusions: Smoothing VF data with NN and VAE algorithms improves the signal-to-noise ratio for detection of change, results in earlier detection of VF progression, and could help monitor glaucoma progression more effectively in the clinical setting. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

A Bayesian Hierarchical Spatial Longitudinal Model Improves Estimation of Local Macular Rates of Change in Glaucomatous Eyes.

Purpose: Demonstrate that a novel Bayesian hierarchical spatial longitudinal (HSL) model improves estimation of local macular ganglion cell complex (GCC) rates of change compared to simple linear regression (SLR) and a conditional autoregressive (CAR) model. Methods: We analyzed GCC thickness measurements within 49 macular superpixels in 111 eyes (111 patients) with four or more macular optical coherence tomography scans and two or more years of follow-up. We compared superpixel-patient-specific estimates and their posterior variances derived from the latest version of a recently developed Bayesian HSL model, CAR, and SLR. We performed a simulation study to compare the accuracy of intercept and slope estimates in individual superpixels. Results: HSL identified a significantly higher proportion of significant negative slopes in 13/49 superpixels and a significantly lower proportion of significant positive slopes in 21/49 superpixels than SLR. In the simulation study, the median (tenth, ninetieth percentile) ratio of mean squared error of SLR [CAR] over HSL for intercepts and slopes were 1.91 (1.23, 2.75) [1.51 (1.05, 2.20)] and 3.25 (1.40, 10.14) [2.36 (1.17, 5.56)], respectively. Conclusions: A novel Bayesian HSL model improves estimation accuracy of patient-specific local GCC rates of change. The proposed model is more than twice as efficient as SLR for estimating superpixel-patient slopes and identifies a higher proportion of deteriorating superpixels than SLR while minimizing false-positive detection rates. Translational Relevance: The proposed HSL model can be used to model macular structural measurements to detect individual glaucoma progression earlier and more efficiently in clinical and research settings.

Detecting Fast Progressors: Comparing a Bayesian Longitudinal Model to Linear Regression for Detecting Structural Changes in Glaucoma.

PURPOSE: Demonstrate that a novel Bayesian hierarchical spatial longitudinal (HSL) model identifies macular superpixels with rapidly deteriorating ganglion cell complex (GCC) thickness more efficiently than simple linear regression (SLR). DESIGN: Prospective cohort study. SETTING: Tertiary Glaucoma Center. SUBJECTS: One hundred eleven eyes (111 patients) with moderate to severe glaucoma at baseline and ≥4 macular optical coherence tomography scans and ≥2 years of follow-up. OBSERVATION PROCEDURE: Superpixel-patient-specific GCC slopes and their posterior variances in 49 superpixels were derived from our latest Bayesian HSL model and Bayesian SLR. A simulation cohort was created with known intercepts, slopes, and residual variances in individual superpixels. MAIN OUTCOME MEASURES: We compared HSL and SLR in the fastest progressing deciles on (1) proportion of superpixels identified as significantly progressing in the simulation study and compared to SLR slopes in cohort data; (2) root mean square error (RMSE), and SLR/HSL RMSE ratios. RESULTS: Cohort- In the fastest decile of slopes per SLR, 77% and 80% of superpixels progressed significantly according to SLR and HSL, respectively. The SLR/HSL posterior SD ratio had a median of 1.83, with 90% of ratios favoring HSL. Simulation- HSL identified 89% significant negative slopes in the fastest progressing decile vs 64% for SLR. SLR/HSL RMSE ratio was 1.36 for the fastest decile of slopes, with 83% of RMSE ratios favoring HSL. CONCLUSION: The Bayesian HSL model improves the estimation efficiency of local GCC rates of change regardless of underlying true rates of change, particularly in fast progressors.

An Artificial Intelligence Enabled System for Retinal Nerve Fiber Layer Thickness Damage Severity Staging.

Purpose: To develop an objective glaucoma damage severity classification system based on OCT-derived retinal nerve fiber layer (RNFL) thickness measurements. Design: Algorithm development for RNFL damage severity classification based on multicenter OCT data. Subjects and Participants: A total of 6561 circumpapillary RNFL profiles from 2269 eyes of 1171 subjects to develop models, and 2505 RNFL profiles from 1099 eyes of 900 subjects to validate models. Methods: We developed an unsupervised k-means model to identify clusters of eyes with similar RNFL thickness profiles. We annotated the clusters based on their respective global RNFL thickness. We computed the optimal global RNFL thickness thresholds that discriminated different severity levels based on Bayes' minimum error principle. We validated the proposed pipeline based on an independent validation dataset with 2505 RNFL profiles from 1099 eyes of 900 subjects. Main Outcome Measures: Accuracy, area under the receiver operating characteristic curve, and confusion matrix. Results: The k-means clustering discovered 4 clusters with 1382, 1613, 1727, and 1839 samples with mean (standard deviation) global RNFL thickness of 58.3 (8.9) µm, 78.9 (6.7) µm, 87.7 (8.2) µm, and 101.5 (7.9) µm. The Bayes' minimum error classifier identified optimal global RNFL values of > 95 µm, 86 to 95 µm, 70 to 85 µm, and < 70 µm for discriminating normal eyes and eyes at the early, moderate, and advanced stages of RNFL thickness loss, respectively. About 4% of normal eyes and 98% of eyes with advanced RNFL loss had either global, or ≥ 1 quadrant, RNFL thickness outside of normal limits provided by the OCT instrument. Conclusions: Unsupervised machine learning discovered that the optimal RNFL thresholds for separating normal eyes and eyes with early, moderate, and advanced RNFL loss were 95 µm, 85 µm, and 70 µm, respectively. This RNFL loss classification system is unbiased as there was no preassumption or human expert intervention in the development process. Additionally, it is objective, easy to use, and consistent, which may augment glaucoma research and day-to-day clinical practice. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Prediction of Central Visual Field Measures From Macular OCT Volume Scans With Deep Learning.

Purpose: Predict central 10° global and local visual field (VF) measurements from macular optical coherence tomography (OCT) volume scans with deep learning (DL). Methods: This study included 1121 OCT volume scans and 10-2 VFs from 289 eyes (257 patients). Macular scans were used to estimate 10-2 VF mean deviation (MD), threshold sensitivities (TS), and total deviation (TD) values at 68 locations. A three-dimensional (3D) convolutional neural network based on the 3D DenseNet121 architecture was used for prediction. We compared DL predictions to those from baseline linear models. We carried out 10-fold stratified cross-validation to optimize generalizability. The performance of the DL and baseline models was compared based on correlations between ground truth and predicted VF measures and mean absolute error (MAE; ground truth - predicted values). Results: Average (SD) MD was -9.3 (7.7) dB. Average (SD) correlations between predicted and ground truth MD and MD MAE were 0.74 (0.09) and 3.5 (0.4) dB, respectively. Estimation accuracy deteriorated with worsening MD. Average (SD) Pearson correlations between predicted and ground truth TS and MAEs for DL and baseline model were 0.71 (0.05) and 0.52 (0.05) (P < 0.001) and 6.5 (0.6) and 7.5 (0.5) dB (P < 0.001), respectively. For TD, correlation (SD) and MAE (SD) for DL and baseline models were 0.69 (0.02) and 0.48 (0.05) (P < 0.001) and 6.1 (0.5) and 7.8 (0.5) dB (P < 0.001), respectively. Conclusions: Macular OCT volume scans can be used to predict global central VF parameters with clinically relevant accuracy. Translational Relevance: Macular OCT imaging may be used to confirm and supplement central VF findings using deep learning.

Prediction of visual field progression with serial optic disc photographs using deep learning.

AIM: We tested the hypothesis that visual field (VF) progression can be predicted with a deep learning model based on longitudinal pairs of optic disc photographs (ODP) acquired at earlier time points during follow-up. METHODS: 3919 eyes (2259 patients) with ≥2 ODPs at least 2 years apart, and ≥5 24-2 VF exams spanning ≥3 years of follow-up were included. Serial VF mean deviation (MD) rates of change were estimated starting at the fifth visit and subsequently by adding visits until final visit. VF progression was defined as a statistically significant negative slope at two consecutive visits and final visit. We built a twin-neural network with ResNet50-backbone. A pair of ODPs acquired up to a year before the VF progression date or the last VF in non-progressing eyes were included as input. Primary outcome measures were area under the receiver operating characteristic curve (AUC) and model accuracy. RESULTS: The average (SD) follow-up time and baseline VF MD were 8.1 (4.8) years and -3.3 (4.9) dB, respectively. VF progression was identified in 761 eyes (19%). The median (IQR) time to progression in progressing eyes was 7.3 (4.5-11.1) years. The AUC and accuracy for predicting VF progression were 0.862 (0.812-0.913) and 80.0% (73.9%-84.6%). When only fast-progressing eyes were considered (MD rate < -1.0 dB/year), AUC increased to 0.926 (0.857-0.994). CONCLUSIONS: A deep learning model can predict subsequent glaucoma progression from longitudinal ODPs with clinically relevant accuracy. This model may be implemented, after validation, for predicting glaucoma progression in the clinical setting.

Comparing Rates of Change in Moderate to Advanced Glaucoma: Retinal Nerve Fiber Layer Versus Bruch Membrane Opening-Minimum Rim Width.

PURPOSE: To compare rates of change (RoC) of peripapillary retinal nerve fiber layer (RNFL) and Bruch membrane opening-based minimum rim width (BMO-MRW) thickness in moderate-to-advanced glaucoma. DESIGN: Prospective cohort study. METHODS: Longitudinal optical coherence tomography (OCT) optic nerve head volume scans of 113 eyes of 113 glaucoma patients with moderate-to-advanced or central damage were exported. This study estimated and compared global and sectoral RoC with linear mixed effects models and simple linear regression (SLR) of RNFL and BMO-MRW thickness. Permutation analyses were used to test significance of RoC in the SLR model. It also compared longitudinal signal-to-noise ratios (LSNR) defined as RoC divided by residual standard deviation (SD) between the two groups. RESULTS: Mean (SD) follow-up and median (IQR) OCT scan sessions were 5.2 (1.3) years and 10 (8-11), respectively. Baseline average (SD) visual field mean deviation was -9.2 (5.8) dB. Based on SLR, a higher proportion of significant negative RNFL RoC was observed compared to BMO-MRW in the inferotemporal (35% vs 20%; P = .015) and inferonasal (42% vs 17%; P < .001) sectors. Permutation analyses also demonstrated a higher proportion of worsening RNFL RoC than BMO-MRW in the inferotemporal (P = .026) and inferonasal (P < .001) sectors along with overall lower positive RoC. Longitudinal signal-to-noise ratios for RNFL were significantly more negative than for BMO-MRW globally, and in the inferotemporal, inferonasal, and superonasal sectors (P ≤ .01). CONCLUSIONS: Longitudinal RNFL OCT measurements are more likely to detect structural change and demonstrate better LSNR compared with BMO-MRW in eyes with central or moderate-to-advanced glaucoma damage at baseline.

Association of Blood Pressure With Rates of Macular Ganglion Cell Complex Thinning in Patients With Glaucoma.

Importance: There are scarce data on the association of blood pressure measures with subsequent macular structural rates of change in patients with glaucoma. Objective: To investigate the association of baseline blood pressure measures with rates of change of the macular ganglion cell complex in patients with central or moderate to advanced glaucoma damage at baseline. Design, Setting, and Participants: This prospective cohort study, conducted from August 2021 to August 2022, used data from patients in the Advanced Glaucoma Progression Study at the University of California, Los Angeles. Participants were between 39 and 80 years of age and had more than 4 macular imaging tests and 2 or more years of follow-up. Exposures: A diagnosis of glaucoma with either central damage or a visual field mean deviation worse than -6 dB. Main Outcomes and Measures: The main outcome was the association of blood pressure measures with ganglion cell complex rates of change. Macular ganglion cell complex thickness rates of change were estimated with a bayesian hierarchical model. This model included relevant demographic and clinical factors. Blood pressure measures, intraocular pressure, and their interactions were added to the model to assess the association of baseline blood pressure measures with global ganglion cell complex rates of change. Results: The cohort included 105 eyes from 105 participants. The mean (SD) age, 10-2 visual field mean deviation, and follow-up time were 66.9 (8.5) years, -8.3 (5.3) dB, and 3.6 (0.4) years, respectively, and 67 patients (63.8%) were female. The racial and ethnic makeup of the cohort was 15 African American (14.3%), 23 Asian (21.9%), 12 Hispanic (11.4%), and 55 White (52.4%) individuals based on patient self-report. In multivariable analyses, female sex, history of taking blood pressure medications, higher intraocular pressure, thicker central corneal thickness, shorter axial length, higher contrast sensitivity at 12 cycles per degree, and higher baseline 10-2 visual field mean deviation were associated with faster ganglion cell complex thinning. Lower diastolic blood pressure was associated with faster rates of ganglion cell complex thinning at higher intraocular pressures. For intraocular pressures of 8 and of 16 mm Hg (10% and 90% quantiles, respectively), every 10 mm Hg-lower increment of diastolic blood pressure was associated with 0.011 µm/y slower and -0.130 µm/y faster rates of ganglion cell complex thinning, respectively. Conclusions and Relevance: In this cohort study, a combination of lower diastolic blood pressure and higher intraocular pressure at baseline was associated with faster rates of ganglion cell complex thinning. These findings support consideration of evaluating and addressing diastolic blood pressure as a therapeutic measure in patients with glaucoma if supported by appropriate clinical trials.

DDLSNet: A Novel Deep Learning-Based System for Grading Funduscopic Images for Glaucomatous Damage.

Purpose: To report an image analysis pipeline, DDLSNet, consisting of a rim segmentation (RimNet) branch and a disc size classification (DiscNet) branch to automate estimation of the disc damage likelihood scale (DDLS). Design: Retrospective observational. Participants: RimNet and DiscNet were developed with 1208 and 11 536 optic disc photographs (ODPs), respectively. DDLSNet performance was evaluated on 120 ODPs from the RimNet test set, for which the DDLS scores were graded by clinicians. Reproducibility was evaluated on a group of 781 eyes, each with 2 ODPs taken within 4 years apart. Methods: Disc damage likelihood scale calculation requires estimation of optic disc size, provided by DiscNet (VGG19 network), and the minimum rim-to-disc ratio (mRDR) or absent rim width (ARW), provided by RimNet (InceptionV3/LinkNet segmentation model). To build RimNet's dataset, glaucoma specialists marked optic disc rim and cup boundaries on ODPs. The "ground truth" mRDR or ARW was calculated. For DiscNet's dataset, corresponding OCT images provided "ground truth" disc size. Optic disc photographs were split into 80/10/10 for training, validation, and testing, respectively, for RimNet and DiscNet. DDLSNet estimation was tested against manual grading of DDLS by clinicians with the average score used as "ground truth." Reproducibility of DDLSNet grading was evaluated by repeating DDLS estimation on a dataset of nonprogressing paired ODPs taken at separate times. Main Outcome Measures: The main outcome measure was a weighted kappa score between clinicians and the DDLSNet pipeline with agreement defined as ± 1 DDLS score difference. Results: RimNet achieved an mRDR mean absolute error (MAE) of 0.04 (± 0.03) and an ARW MAE of 48.9 (± 35.9) degrees when compared to clinician segmentations. DiscNet achieved 73% (95% confidence interval [CI]: 70%, 75%) classification accuracy. DDLSNet achieved an average weighted kappa agreement of 0.54 (95% CI: 0.40, 0.68) compared to clinicians. Average interclinician agreement was 0.52 (95% CI: 0.49, 0.56). Reproducibility testing demonstrated that 96% of ODP pairs had a difference of ≤ 1 DDLS score. Conclusions: DDLSNet achieved moderate agreement with clinicians for DDLS grading. This novel approach illustrates the feasibility of automated ODP grading for assessing glaucoma severity. Further improvements may be achieved by increasing the number of incomplete rims sample size, expanding the hyperparameter search, and increasing the agreement of clinicians grading ODPs.

Comparison of Ganglion Cell Layer and Ganglion Cell/Inner Plexiform Layer Measures for Detection of Early Glaucoma.

PURPOSE: To test the hypothesis that macular ganglion cell layer (GCL) measurements detect early glaucoma with higher accuracy than ganglion cell/inner plexiform layer (GCIPL) thickness measurements. DESIGN: Cross-sectional study. PARTICIPANTS: The first cohort included 58 glaucomatous eyes with visual field mean deviation (MD) ≥ -6 dB and 125 normal eyes. The second cohort included 72 glaucomatous and 73 normal/glaucoma suspect (GS) eyes with scans able to create GCL/GCIPL deviation maps. METHODS: In the first cohort, 8 × 8 GCL and GCIPL grids were exported and 5 superior and inferior sectors were defined. Global and sectoral GCL and GCIPL measures were used to predict glaucoma. In the second cohort, proportions of scan areas with abnormal (< 5% and < 1% cutoffs) and supernormal (> 95% and > 99% cutoffs) thicknesses on deviation maps were calculated. The extents of GCL and GCIPL abnormal areas were used to predict glaucoma. MAIN OUTCOME MEASURES: Extents of abnormal GCL/GCIPL regions and areas under receiver operating characteristic curves (AUROC) for prediction of glaucoma were compared between GCL or GCIPL measures. RESULTS: The average ± standard deviation MDs were -3.7 ± 1.6 dB and -2.7 ± 1.8 dB in glaucomatous eyes in the first and second cohorts, respectively. Global GCIPL thickness measures (central 18° × 18° macular region) performed better than GCL for early detection of glaucoma (AUROC, 0.928 vs. 0.884, respectively; P = 0.004). Superior and inferior sector 3 thickness measures provided the best discrimination with both GCL and GCIPL (inferior GCL AUROC, 0.860 vs. GCIPL AUROC, 0.916 [P = 0.001]; superior GCL AUROC, 0.916 vs. GCIPL AUROC, 0.900 [P = 0.24]). The extents of abnormal GCL regions at a 1% cutoff in the central elliptical area were 17.5 ± 22.2% and 6.4 ± 10.8% in glaucomatous and normal/GS eyes, respectively, versus 17.0 ± 22.2% and 5.7 ± 10.5%, respectively, for GCIPL (P = 0.06 for GCL and 0.002 for GCIPL). The extents of GCL and GCIPL supernormal regions were mostly similar in glaucomatous and normal eyes. The best performance for prediction of glaucoma in the second cohort was detected at a P value of < 1% within the entire scan for both GCL and GCIPL (AUC, 0.681 vs. 0.668, respectively; P = 0.29). CONCLUSIONS: Macular GCL and GCIPL thicknesses are equivalent for identifying early glaucoma with current OCT technology. This is likely explained by limitations of inner macular layer segmentation and concurrent changes within the inner plexiform layer in early glaucoma.

Combined phacoviscocanalostomy versus phacoemulsification alone in patients with coexisting cataract and mild-to-moderate open-angle glaucoma; a randomized-controlled trial.

BACKGROUND/OBJECTIVE: Management of concomitant cataract and glaucoma depends on the stage of glaucoma and the patient's situation. There are different surgical options for handling visually significant cataract and mild-to-moderate open-angle glaucoma (OAG). We aimed to compare the one-year results of phacoemulsification alone versus phacoviscocanalostomy in these patients. SUBJECTS/METHODS: This was a parallel-arm, single-masked, randomized-controlled trial, conducted at Farabi Eye Hospital, Tehran, Iran between January 2016 and January 2018. We enrolled 89 eyes from 89 patients with mild-to-moderate primary OAG or pseudoexfoliative glaucoma (PEXG) with visually significant age-related cataract. They randomly underwent phacoemulsification alone (n = 44) or combined phaco-viscocanalostomy (n = 45). All patients had a 12-month follow-up period, and the mean intraocular pressure (IOP), the number of antiglaucoma medications, and complete and qualified success rates were compared. RESULTS: After the 1st and 3rd months, the mean IOP showed significantly decreased in the phaco-visco group compared to the phaco group (P < 0001 and P = 0.004, respectively), but it was not statistically significant at 6th and 12th months (P = 0.540 and P = 0.530). The need for antiglaucoma medication and the complete and qualified success rates were significantly in favour of the phaco-visco group in all postoperative visits (P < 0.05). CONCLUSIONS: Although both phacoemulsification alone and phacoviscocanalostomy procedures can be considered for patients with mild-to-moderate OAG, we found better success rates using phacoviscocanalostomy. Therefore, if the surgeon is an expert in performing this technique, this non-penetrating procedure can be applied in patients with visually significant cataract and earlier stages of OAG, especially in patients with PEXG.

Measurement of the Inner Macular Layers for Monitoring of Glaucoma: Confounding Effects of Age-Related Macular Degeneration.

OBJECTIVE: To investigate the confounding effect of nonexudative age-related macular degeneration (AMD), specifically drusen and outer retinal atrophy, on the architecture and automated segmentation of the inner retinal layers as measured with OCT. DESIGN: Observational cross-sectional study. SUBJECTS: Two hundred sixty-three consecutive eyes with nonexudative AMD were identified through a retrospective chart review. Exclusion criteria were a diagnosis of glaucoma or glaucoma suspect, other retinal pathology affecting the macula, axial length > 26.5 mm or spherical equivalent less than -6 diopters, any other optic nerve or neurologic disorders, or poor image quality. METHODS: Drusen were automatically segmented on macular OCT B-scans with a publicly available and validated deep learning approach. Automated segmentation of the inner plexiform layer (IPL)/inner nuclear layer (INL) boundary was carried out with the device's proprietary software. MAIN OUTCOME MEASURES: Quality of segmentation of the IPL/INL boundary as a function of drusen size and presence of inner retinal layer displacement in the area of macular pathology (drusen or atrophy). RESULTS: One hundred twenty-five eyes (65 patients) met the inclusion criteria. Drusen size varied between 16 and 272 µm (mean, 118 µm). Automated segmentation had a 22% chance of failure if the drusen height was between 145 and 185 µm and was most likely to fail with drusen heights above 185 µm. When drusen height was normalized by total retinal thickness, segmentation failed 36% of the time when the drusen to total retinal thickness ratio was 0.45 or above. Images were likely to show displacement of inner retinal layers with drusen heights above 176 µm and a normalized drusen height ratio of 0.5 or higher. Eighty-seven percent of images with outer retinal atrophy displayed incorrect segmentation. CONCLUSIONS: Outer retinal diseases can alter the retinal topography and affect the segmentation accuracy of the inner retinal layers. Large drusen may cause segmentation error and compression of the inner macular layers. Geographic atrophy confounds automated segmentation in a high proportion of eyes. Clinicians should be cognizant of the effects of outer retinal disease on the inner retinal layer measurements when interpreting the results of macular OCT imaging in patients with glaucoma.

Risk factors for microcystic macular oedema in glaucoma.

BACKGROUND/AIMS: To identify clinical characteristics and factors associated with microcystic macular edema (MME) in patients with primary open-angle glaucoma (POAG). METHODS: We included 315 POAG eyes between 2010 and 2019 with good-quality macular volume scans that had reliable visual fields (VF) available within 6 months in this observational retrospective cohort study. Eyes with retinal pathologies except for epiretinal membrane (ERM) were excluded. The inner nuclear layer was qualitatively assessed for the presence of MME. Global mean deviation (MD) and Visual Field Index (VFI) decay rates, superior and inferior MD rates and pointwise total deviation rates of change were estimated with linear regression. Logistic regression was performed to identify baseline factors associated with the presence of MME and to determine whether MME is associated with progressive VF loss. RESULTS: 25 out of 315 eyes (7.9%) demonstrated MME. The average (±SD) age and MD in eyes with and without MME was 57.2 (±8.7) versus 62.0 (±9.9) years (p=0.02) and -9.8 (±5.7) versus -4.9 (±5.3) dB (p<0.001), respectively. Worse global MD at baseline (p=0.001) and younger age (p=0.02) were associated with presence of MME. ERM was not associated with the presence of MME (p=0.84) in this cohort. MME was not associated with MD and VFI decay rates (p>0.49). CONCLUSIONS: More severe glaucoma and younger age were associated with MME. MME was not associated with faster global VF decay in this cohort. MME may confound monitoring of glaucoma with full macular thickness.

RimNet: A Deep Neural Network Pipeline for Automated Identification of the Optic Disc Rim.

Purpose: Accurate neural rim measurement based on optic disc imaging is important to glaucoma severity grading and often performed by trained glaucoma specialists. We aim to improve upon existing automated tools by building a fully automated system (RimNet) for direct rim identification in glaucomatous eyes and measurement of the minimum rim-to-disc ratio (mRDR) in intact rims, the angle of absent rim width (ARW) in incomplete rims, and the rim-to-disc-area ratio (RDAR) with the goal of optic disc damage grading. Design: Retrospective cross sectional study. Participants: One thousand and twenty-eight optic disc photographs with evidence of glaucomatous optic nerve damage from 1021 eyes of 903 patients with any form of primary glaucoma were included. The mean age was 63.7 (± 14.9) yrs. The average mean deviation of visual fields was -8.03 (± 8.59). Methods: The images were required to be of adequate quality, have signs of glaucomatous damage, and be free of significant concurrent pathology as independently determined by glaucoma specialists. Rim and optic cup masks for each image were manually delineated by glaucoma specialists. The database was randomly split into 80/10/10 for training, validation, and testing, respectively. RimNet consists of a deep learning rim and cup segmentation model, a computer vision mRDR measurement tool for intact rims, and an ARW measurement tool for incomplete rims. The mRDR is calculated at the thinnest rim section while ARW is calculated in regions of total rim loss. The RDAR was also calculated. Evaluation on the Drishti-GS dataset provided external validation (Sivaswamy 2015). Main Outcome Measures: Median Absolute Error (MAE) between glaucoma specialists and RimNet for mRDR and ARW. Results: On the test set, RimNet achieved a mRDR MAE of 0.03 (0.05), ARW MAE of 31 (89)°, and an RDAR MAE of 0.09 (0.10). On the Drishti-GS dataset, an mRDR MAE of 0.03 (0.04) and an mRDAR MAE of 0.09 (0.10) was observed. Conclusions: RimNet demonstrated acceptably accurate rim segmentation and mRDR and ARW measurements. The fully automated algorithm presented here would be a valuable component in an automated mRDR-based glaucoma grading system. Further improvements could be made by improving identification and segmentation performance on incomplete rims and expanding the number and variety of glaucomatous training images.

Comparison of Ganglion Cell Layer and Inner Plexiform Layer Rates of Change in Suspected and Established Glaucoma.

PURPOSE: We compared ganglion cell layer (GCL) and inner plexiform layer (IPL) rates of change (RoC) in patients with glaucoma suspect (GS) and established glaucoma (EG) to test the hypothesis that IPL thickness changes would occur earlier than GCL changes in eyes with early damage. DESIGN: Prospective, cohort study. METHODS: A total of 64 GS eyes (46 patients) and 112 EG eyes (112 patients) with ≥2 years of follow-up and ≥3 macular optical coherence tomography scans were included. GCL and IPL superpixel thickness measurements were exported. A Bayesian hierarchical model with random intercepts/slopes and random residual variances was fitted to estimate RoC in individual superpixels. Normalized RoC and proportions of superpixels with significantly negative and positive GCL and IPL RoC were compared within the groups. RESULTS: The average (SD) follow-up time and number of scans were 3.5 (0.7) years and 4.2 (1.0), respectively, in the GS group and 3.6 (0.4) years and 7.3 (1.1) in the EG group. Mean (SD) normalized RoC was faster for GCL than IPL (-0.69 [0.05] vs -0.33 [0.04]) in the GS group, whereas it was faster for IPL (-0.47 [0.03] vs -0.28 [0.02]) in EG eyes. GCL RoC were significantly negative in 24 of 36 superpixels compared with 8 of 36 for IPL (P < .001) in GS eyes. In the EG group, 23 of 36 superpixels had significant negative IPL RoC compared with 13 of 36 superpixels for GCL (P = .006). CONCLUSIONS: GCL thickness is more likely to demonstrate change over time compared with IPL in glaucoma suspects. There is no evidence of preferential IPL thinning in eyes with suspected early glaucoma damage.

Multivariate Longitudinal Modeling of Macular Ganglion Cell Complex: Spatiotemporal Correlations and Patterns of Longitudinal Change.

Purpose: To investigate spatiotemporal correlations among ganglion cell complex (GCC) superpixel thickness measurements and explore underlying patterns of longitudinal change across the macular region. Design: Longitudinal cohort study. Subjects: One hundred eleven eyes from 111 subjects from the Advanced Glaucoma Progression Study with ≥ 4 visits and ≥ 2 years of follow-up. Methods: We further developed our proposed Bayesian hierarchical model for studying longitudinal GCC thickness changes across macular superpixels in a cohort of glaucoma patients. Global priors were introduced for macular superpixel parameters to combine data across superpixels and better estimate population slopes and intercepts. Main Outcome Measures: Bayesian residual analysis to inspect cross-superpixel correlations for subject random effects and residuals. Principal component analysis (PCA) to explore underlying patterns of longitudinal macular change. Results: Average (standard deviation [SD]) follow-up and baseline 10-2 visual field mean deviation were 3.6 (0.4) years and -8.9 (5.9) dB, respectively. Superpixel-level random effects and residuals had the greatest correlations with nearest neighbors; correlations were higher in the superior than in the inferior region and strongest among random intercepts, followed by random slopes, residuals, and residual SDs. PCA of random intercepts showed a first large principal component (PC) across superpixels that approximated a global intercept, a second PC that contrasted the superior and inferior macula, and a third PC, contrasting inner and nasal superpixels with temporal and peripheral superpixels. PCs for slopes, residual SDs, and residuals were remarkably similar to those of random intercepts. Conclusions: Introduction of cross-superpixel random intercepts and slopes is expected to improve estimation of population and subject parameters. Further model enhancement may be possible by including cross-superpixel random effects and correlations to address spatiotemporal relationships in longitudinal data sets.

A Metascore of Multiple Imaging Methods to Measure Long-Term Glaucoma Structural Progression.

Purpose: To develop a structural metascore (SMS) that combines measurements from different devices and expresses them on a single scale to facilitate their long-term analysis. Methods: Three structural measurements (Heidelberg Retina Tomograph II [HRT] rim area, HD-Cirrus optical coherence tomography [OCT] average retinal nerve fiber layer [RNFL] thickness, Spectralis OCT RNFL global thickness) were normalized on a scale of 0 to 100 and converted to a reference value. The resultant metascores were plotted against time. SMS performance was evaluated to predict future values (internal validation), and correlations between the average grades assigned by three clinicians were compared with the SMS slopes (external validation). Results: The linear regression fit with the variance approach, and adjustment to a Spectralis equivalent was the best-performing approach; this was denominated metascore. Plots were created for 3416 eyes of 1824 patients. The average baseline age (± standard deviation) was 69.8 (±13.9), mean follow-up was 11.6 (±4.7) years, and mean number of structural scans per eye was 10.0 (±4.7). The mean numbers of scans per device were 3.8 (±2.5), 5.0 (±2.9), and 1.3 (±3.0) for HRT, Cirrus, and Spectralis, respectively. The metascore slopes' median was -0.3 (interquartile range 1.1). Correlations between the average grades assigned by the three clinicians and the metascore slopes were -0.51, -0.49, and -0.69 for the first (structural measurement printouts alone), second (metascore plots alone), and third (printouts + metascore plots) series of gradings, respectively. The average absolute predictive ability was 7.63/100 (whereas 100 = entire normalized scale). Conclusions: We report a method that converts Cirrus global RNFL and HRT global rim area normalized measurements to Spectralis global RNFL equivalent values to facilitate long-term structural follow-up. Translational Relevance: Because glaucoma changes usually occur slowly, patients are often examined with different instruments during their follow-up, a method that "unifies" structural measurements provided by different devices, which could assist patients' longitudinal structural follow-up.

Central Macular Topographic and Volumetric Measures: New Biomarkers for Detection of Glaucoma.

Purpose: To test the hypothesis that newly developed shape measures using optical coherence tomography (OCT) macular volume scans can discriminate patients with perimetric glaucoma from healthy subjects. Methods: OCT structural measures defining macular topography and volume were recently developed based on cubic Bézier curves. We exported macular volume scans from 135 eyes with glaucoma (133 patients) and 155 healthy eyes (85 subjects) and estimated global and quadrant-based measures. The best subset of measures to predict glaucoma was explored with a gradient boost model (GBM) with subsequent logistic regression. Accuracy and area under receiver operating curves (AUC) were the primary metrics. In addition, we separately investigated model performance in 66 eyes with mild glaucoma (mean deviation ≥ -6 dB). Results: Average (±SD) 24-2 mean deviation was -8.2 (±6.1) dB in eyes with glaucoma. The main predictive measures for glaucoma were temporal inferior rim height, nasal inferior pit volume, and temporal inferior pit depth. Lower values for these measures predicted higher risk of glaucoma. Sensitivity, specificity, and AUC for discriminating between healthy and glaucoma eyes were 81.5% (95% CI = 76.6-91.9%), 89.7% (95% CI = 78.7-94.2%), and 0.915 (95% CI = 0.882-0.948), respectively. Corresponding metrics for mild glaucoma were 84.8% (95% CI = 72.1%-95.5%), 85.8% (95% CI = 87.1%-97.4%), and 0.913 (95% CI = 0.867-0.958), respectively. Conclusions: Novel macular shape biomarkers detect early glaucoma with clinically relevant performance. Such biomarkers do not depend on intraretinal segmentation accuracy and may be helpful in eyes with suboptimal macular segmentation. Translational Relevance: Macular shape biomarkers provide valuable information for detection of early glaucoma and may provide additional information beyond thickness measurements.