Deep Learning Identifies High-Quality Fundus Photographs and Increases Accuracy in Automated Primary Open Angle Glaucoma Detection.

Purpose: To develop and evaluate a deep learning (DL) model to assess fundus photograph quality, and quantitatively measure its impact on automated POAG detection in independent study populations. Methods: Image quality ground truth was determined by manual review of 2815 fundus photographs of healthy and POAG eyes from the Diagnostic Innovations in Glaucoma Study and African Descent and Glaucoma Evaluation Study (DIGS/ADAGES), as well as 11,350 from the Ocular Hypertension Treatment Study (OHTS). Human experts assessed a photograph as high quality if of sufficient quality to determine POAG status and poor quality if not. A DL quality model was trained on photographs from DIGS/ADAGES and tested on OHTS. The effect of DL quality assessment on DL POAG detection was measured using area under the receiver operating characteristic (AUROC). Results: The DL quality model yielded an AUROC of 0.97 for differentiating between high- and low-quality photographs; qualitative human review affirmed high model performance. Diagnostic accuracy of the DL POAG model was significantly greater (P < 0.001) in good (AUROC, 0.87; 95% CI, 0.80-0.92) compared with poor quality photographs (AUROC, 0.77; 95% CI, 0.67-0.88). Conclusions: The DL quality model was able to accurately assess fundus photograph quality. Using automated quality assessment to filter out low-quality photographs increased the accuracy of a DL POAG detection model. Translational Relevance: Incorporating DL quality assessment into automated review of fundus photographs can help to decrease the burden of manual review and improve accuracy for automated DL POAG detection.

Time to Glaucoma Progression Detection by Optical Coherence Tomography in Individuals of African and European Descents.

PURPOSE: To examine the time to detectable retinal nerve fiber layer thickness (RNFLT) progression by optical coherence tomography (OCT) among glaucoma patients of African descent (AD) and European descent (ED). DESIGN: Retrospective cohort study. METHODS: AD and ED glaucoma eyes from the Diagnostic Innovations in Glaucoma Study (DIGS)/African Descent and Glaucoma Evaluation Study (ADAGES) with ≥2 years/4 visits of optic nerve head RNFLT measurements were included after homogenization on age, diagnosis, and baseline visual field (VF) measurement. RNFLT variability estimates based on linear mixed-effects models were used to simulate longitudinal RNFLT data for both races. Times to trend-based RNFLT progression detection were calculated under standardized scenarios (same RNFLT baseline/thinning rates for both races) and real-world scenarios (AD and ED cohort-specific RNFLT baseline/thinning rates). RESULTS: We included 332 and 542 eyes (216 and 317 participants) of AD and ED, respectively. In standardized scenarios, the time to detect RNFLT progression appeared to be similar (difference, <0.2 years) for AD and ED across different assumed RNFLT thinning rates/baseline. In real-world scenarios, compared to ED, AD had a faster RNFLT thinning rate (-0.8 vs -0.6 µm/y) and thicker baseline RNFLT (84.6 vs 81.8 µm). With a faster thinning rate, the mean (SD) time to progression detection was shorter in AD (4.8 [2.0] vs ED: 5.4 [2.4] years), and the 5-year progression rate appeared to be higher (AD: 59% vs ED: 47%). CONCLUSIONS: Time to progression detection was similar for both races when assuming identical RNFLT baseline/thinning rates, and shorter in AD eyes under real-world simulation when AD had faster RNFLT thinning. In contrast to prior results on VF, which detected progression later in AD eyes than in ED eyes, OCT may detect progression more consistently across these races.

One-Vote Veto: Semi-Supervised Learning for Low-Shot Glaucoma Diagnosis.

Convolutional neural networks (CNNs) are a promising technique for automated glaucoma diagnosis from images of the fundus, and these images are routinely acquired as part of an ophthalmic exam. Nevertheless, CNNs typically require a large amount of well-labeled data for training, which may not be available in many biomedical image classification applications, especially when diseases are rare and where labeling by experts is costly. This article makes two contributions to address this issue: 1) It extends the conventional Siamese network and introduces a training method for low-shot learning when labeled data are limited and imbalanced, and 2) it introduces a novel semi-supervised learning strategy that uses additional unlabeled training data to achieve greater accuracy. Our proposed multi-task Siamese network (MTSN) can employ any backbone CNN, and we demonstrate with four backbone CNNs that its accuracy with limited training data approaches the accuracy of backbone CNNs trained with a dataset that is 50 times larger. We also introduce One-Vote Veto (OVV) self-training, a semi-supervised learning strategy that is designed specifically for MTSNs. By taking both self-predictions and contrastive predictions of the unlabeled training data into account, OVV self-training provides additional pseudo labels for fine-tuning a pre-trained MTSN. Using a large (imbalanced) dataset with 66,715 fundus photographs acquired over 15 years, extensive experimental results demonstrate the effectiveness of low-shot learning with MTSN and semi-supervised learning with OVV self-training. Three additional, smaller clinical datasets of fundus images acquired under different conditions (cameras, instruments, locations, populations) are used to demonstrate the generalizability of the proposed methods.

Detecting Glaucoma from Fundus Photographs Using Deep Learning without Convolutions: Transformer for Improved Generalization.

Purpose: To compare the diagnostic accuracy and explainability of a Vision Transformer deep learning technique, Data-efficient image Transformer (DeiT), and ResNet-50, trained on fundus photographs from the Ocular Hypertension Treatment Study (OHTS) to detect primary open-angle glaucoma (POAG) and identify the salient areas of the photographs most important for each model's decision-making process. Design: Evaluation of a diagnostic technology. Subjects Participants and Controls: Overall 66 715 photographs from 1636 OHTS participants and an additional 5 external datasets of 16 137 photographs of healthy and glaucoma eyes. Methods: Data-efficient image Transformer models were trained to detect 5 ground-truth OHTS POAG classifications: OHTS end point committee POAG determinations because of disc changes (model 1), visual field (VF) changes (model 2), or either disc or VF changes (model 3) and Reading Center determinations based on disc (model 4) and VFs (model 5). The best-performing DeiT models were compared with ResNet-50 models on OHTS and 5 external datasets. Main Outcome Measures: Diagnostic performance was compared using areas under the receiver operating characteristic curve (AUROC) and sensitivities at fixed specificities. The explainability of the DeiT and ResNet-50 models was compared by evaluating the attention maps derived directly from DeiT to 3 gradient-weighted class activation map strategies. Results: Compared with our best-performing ResNet-50 models, the DeiT models demonstrated similar performance on the OHTS test sets for all 5 ground-truth POAG labels; AUROC ranged from 0.82 (model 5) to 0.91 (model 1). Data-efficient image Transformer AUROC was consistently higher than ResNet-50 on the 5 external datasets. For example, AUROC for the main OHTS end point (model 3) was between 0.08 and 0.20 higher in the DeiT than ResNet-50 models. The saliency maps from the DeiT highlight localized areas of the neuroretinal rim, suggesting important rim features for classification. The same maps in the ResNet-50 models show a more diffuse, generalized distribution around the optic disc. Conclusions: Vision Transformers have the potential to improve generalizability and explainability in deep learning models, detecting eye disease and possibly other medical conditions that rely on imaging for clinical diagnosis and management.

Effect of Testing Frequency on the Time to Detect Glaucoma Progression With Optical Coherence Tomography (OCT) and OCT Angiography.

PURPOSE: To determine how the frequency of testing affects the time required to detect statistically significant glaucoma progression for circumpapillary retinal nerve fiber layer (cpRNFL) with optical coherence tomography (OCT) and circumpapillary capillary density (cpCD) with OCT angiography (OCTA). DESIGN: Retrospective, observational cohort study. METHODS: In this longitudinal study, 156 eyes of 98 patients with glaucoma followed up over an average of 3.5 years were enrolled. Participants with 4 or more OCT and OCTA tests were included to measure the longitudinal rates of cpRNFL thickness and cpCD change over time using linear regression. Estimates of variability were then used to re-create real-world cpRNFL and cpCD data by computer simulation to evaluate the time required to detect progression for various loss rates and different testing frequencies. RESULTS: The time required to detect a statistically significant negative cpRNFL and cpCD slope decreased as the testing frequency increased, albeit not proportionally. cpCD detected progression slightly earlier than cpRNFL. Eighty percent of eyes with a cpCD loss of -1%/y were detected after 6.0, 4.2, and 4 years when testing was performed 1, 2, and 3 times per year, respectively. Progression in 80% of eyes with a cpRNFL loss of -1 µm/y was detected after 6.3, 5.0, and 4.2 years, respectively. CONCLUSIONS: cpRNFL and cpCD are comparable in detecting progression. As there were only small changes in the time to detect progression when testing increased from 2 to 3 times per year, testing twice per year may provide sufficient information for detecting progression with either OCT or OCTA in clinical settings.

Comparison of Optic Disc Ovality Index and Rotation Angle Measurements in Myopic Eyes Using Photography and OCT Based Techniques.

Purpose: To compare optic nerve head (ONH) ovality index and rotation angle measurements based on semi-automated delineation of the clinical ONH margin derived from photographs and automated BMO configuration derived from optical coherence tomography (OCT) images in healthy and glaucomatous eyes with high-, mild- and no axial myopia. Methods: One hundred seventy-five healthy and glaucomatous eyes of 146 study participants enrolled in the Diagnostic Innovations in Glaucoma Study (DIGS) with optic disc photographs and Spectralis OCT ONH scans acquired on the same day were stratified by level of axial myopia (non-myopic [n = 56, axial length (AL) <24 mm], mild-myopic [n = 58, AL 24-26 mm] and high-myopic [n = 32, AL >26 mm]. The clinical disc margin of each photograph was manually annotated, and semi-automated measurements were recorded of the ovality index and rotation angle based on a best-fit ellipse generated using ImageJ software. These semi-automated photograph-based measurements were compared to ovality index and rotation angle generated from custom automated BMO-based analysis using segmented OCT ONH volumes. R 2 values from linear mixed effects models were used to describe the associations between semi-automated, photograph-based and automated OCT-based measurements. Results: Average (95% CI) axial length was 23.3 (23.0, 23.3) mm, 24.8 (24.7, 25.0) mm and 26.8 (26.6, 27.0) mm in non-myopic, mild-myopic and high-myopic eyes, respectively (ANOVA, p ≤ 0.001 for all). The R 2 association (95% CI) between semi-automated photograph-based and automated OCT-based assessment of ONH OI for all eyes was [0.26 (0.16, 0.36); p < 0.001]. This association was weakest in non-myopic eyes [0.09 (0.01, 0.26); p = 0.02], followed by mild-myopic eyes [0.13 (0.02, 0.29); p = 0.004] and strongest in high-myopic eyes [0.40 (0.19, 0.60); p < 0.001]. No significant associations were found between photography- and OCT-based assessment of rotation angle with R 2 values ranging from 0.00 (0.00, 0.08) in non-myopic eyes to 0.03 (0.00, 0.21) in high-myopic eyes (all associations p ≥ 0.33). Conclusions: Agreement between photograph-based and automated OCT-based ONH morphology measurements is limited, suggesting that these methods cannot be used interchangeably for characterizing myopic changes in the ONH.

Comparison of the Effects of Latanoprostene Bunod and Timolol on Retinal Blood Vessel Density: A Randomized Clinical Trial.

PURPOSE: To compare the differences in retinal vessel density (VD) between topical administration of latanoprostene bunod (LBN) ophthalmic solution 0.024% and timolol maleate 0.5% in patients with open-angle glaucoma (OAG) or ocular hypertension (OHT) and normal subjects. DESIGN: Randomized, single center, crossover clinical trial. METHODS: Eligible subjects were examined during 6 study visits over 12 weeks. All subjects were randomized in a 1:1 ratio to LBN dosed once daily or timolol dosed twice daily in both eyes (OU) for a duration of 4 weeks each, separated by a 2-week washout period. A comprehensive eye examination OU was performed at each visit. Testing was performed with optical coherence tomography and optical coherence tomography angiography (optic nerve and macula), as well as visual field examination, on the study eye at baseline and before and after each treatment. RESULTS: One eye from each of 50 patients was enrolled (10 healthy patients, 26 patients with OHT, and 14 patients with OAG). After administration of LBN there was significantly increased macular VD (0.76% [0.20%-1.33%], P = 0.009) and a trend in increasing peripapillary VD in patients with OAG and patients with OHT. In contrast, after administration of timolol, there were no differences in macular VD, and a decrease in peripapillary VD only was observed in the nasal inferior sector (-0.56% [-1.08% to -0.03%], P = .04) in patients with OAG and patients with OHT. No change in peripapillary or macular VD was observed in the normal subjects (P > .05 for all). CONCLUSIONS: Topical administration of LBN enhanced macular VD in patients with OAG or patients with OHT. In contrast, timolol administration did not have any effect on VD.

Detecting Glaucoma in the Ocular Hypertension Study Using Deep Learning.

Importance: Automated deep learning (DL) analyses of fundus photographs potentially can reduce the cost and improve the efficiency of reading center assessment of end points in clinical trials. Objective: To investigate the diagnostic accuracy of DL algorithms trained on fundus photographs from the Ocular Hypertension Treatment Study (OHTS) to detect primary open-angle glaucoma (POAG). Design, Setting, and Participants: In this diagnostic study, 1636 OHTS participants from 22 sites with a mean (range) follow-up of 10.7 (0-14.3) years. A total of 66 715 photographs from 3272 eyes were used to train and test a ResNet-50 model to detect the OHTS Endpoint Committee POAG determination based on optic disc (287 eyes, 3502 photographs) and/or visual field (198 eyes, 2300 visual fields) changes. Three independent test sets were used to evaluate the generalizability of the model. Main Outcomes and Measures: Areas under the receiver operating characteristic curve (AUROC) and sensitivities at fixed specificities were calculated to compare model performance. Evaluation of false-positive rates was used to determine whether the DL model detected POAG before the OHTS Endpoint Committee POAG determination. Results: A total of 1147 participants were included in the training set (661 [57.6%] female; mean age, 57.2 years; 95% CI, 56.6-57.8), 167 in the validation set (97 [58.1%] female; mean age, 57.1 years; 95% CI, 55.6-58.7), and 322 in the test set (173 [53.7%] female; mean age, 57.2 years; 95% CI, 56.1-58.2). The DL model achieved an AUROC of 0.88 (95% CI, 0.82-0.92) for the OHTS Endpoint Committee determination of optic disc or VF changes. For the OHTS end points based on optic disc changes or visual field changes, AUROCs were 0.91 (95% CI, 0.88-0.94) and 0.86 (95% CI, 0.76-0.93), respectively. False-positive rates (at 90% specificity) were higher in photographs of eyes that later developed POAG by disc or visual field (27.5% [56 of 204]) compared with eyes that did not develop POAG (11.4% [50 of 440]) during follow-up. The diagnostic accuracy of the DL model developed on the optic disc end point applied to 3 independent data sets was lower, with AUROCs ranging from 0.74 (95% CI, 0.70-0.77) to 0.79 (95% CI, 0.78-0.81). Conclusions and Relevance: The model's high diagnostic accuracy using OHTS photographs suggests that DL has the potential to standardize and automate POAG determination for clinical trials and management. In addition, the higher false-positive rate in early photographs of eyes that later developed POAG suggests that DL models detected POAG in some eyes earlier than the OHTS Endpoint Committee, reflecting the OHTS design that emphasized a high specificity for POAG determination by requiring a clinically significant change from baseline.