Deep Learning-Assisted Detection of Glaucoma Progression in Spectral-Domain OCT.

PURPOSE: To develop and validate a deep learning (DL) model for detection of glaucoma progression using spectral-domain (SD)-OCT measurements of retinal nerve fiber layer (RNFL) thickness. DESIGN: Retrospective cohort study. PARTICIPANTS: A total of 14 034 SD-OCT scans from 816 eyes from 462 individuals. METHODS: A DL convolutional neural network was trained to assess SD-OCT RNFL thickness measurements of 2 visits (a baseline and a follow-up visit) along with time between visits to predict the probability of glaucoma progression. The ground truth was defined by consensus from subjective grading by glaucoma specialists. Diagnostic performance was summarized by the area under the receiver operator characteristic curve (AUC), sensitivity, and specificity, and was compared with conventional trend-based analyses of change. Interval likelihood ratios were calculated to determine the impact of DL model results in changing the post-test probability of progression. MAIN OUTCOME MEASURES: The AUC, sensitivity, and specificity of the DL model. RESULTS: The DL model had an AUC of 0.938 (95% confidence interval [CI], 0.921-0.955), with sensitivity of 87.3% (95% CI, 83.6%-91.6%) and specificity of 86.4% (95% CI, 79.9%-89.6%). When matched for the same specificity, the DL model significantly outperformed trend-based analyses. Likelihood ratios for the DL model were associated with large changes in the probability of progression in the vast majority of SD-OCT tests. CONCLUSIONS: A DL model was able to assess the probability of glaucomatous structural progression from SD-OCT RNFL thickness measurements. The model agreed well with expert judgments and outperformed conventional trend-based analyses of change, while also providing indication of the likely locations of change. FINANCIAL DISCLOSURE(S): Proprietary or commercial disclosure may be found after the references.

Longitudinal visual field variability and the ability to detect glaucoma progression in black and white individuals.

BACKGROUND/AIMS: To investigate racial differences in the variability of longitudinal visual field testing in a 'real-world' clinical population, evaluate how these differences are influenced by socioeconomic status, and estimate the impact of differences in variability on the time to detect visual field progression. METHODS: This retrospective observational cohort study used data from 1103 eyes from 751 White individuals and 428 eyes from 317 black individuals. Linear regression was performed on the standard automated perimetry mean deviation values for each eye over time. The SD of the residuals from the trend lines was calculated and used as a measure of variability for each eye. The association of race with the SD of the residuals was evaluated using a multivariable generalised estimating equation model with an interaction between race and zip code income. Computer simulations were used to estimate the time to detect visual field progression in the two racial groups. RESULTS: Black patients had larger visual field variability over time compared with white patients, even when adjusting for zip code level socioeconomic variables (SD of residuals for Black patients=1.53 dB (95% CI 1.43 to 1.64); for white patients=1.26 dB (95% CI 1.14 to 1.22); mean difference: 0.28 (95% CI 0.15 to 0.41); p<0.001). The difference in visual field variability between black and white patients was greater at lower levels of income and led to a delay in detection of glaucoma progression. CONCLUSION: Black patients had larger visual field variability compared with white patients. This relationship was strongly influenced by socioeconomic status and may partially explain racial disparities in glaucoma outcomes.

Blood Pressure and Glaucomatous Progression in a Large Clinical Population.

PURPOSE: To investigate the effect of systemic arterial blood pressure (BP) on rates of progressive structural damage over time in glaucoma. DESIGN: Retrospective cohort study. PARTICIPANTS: A total of 7501 eyes of 3976 subjects with glaucoma or suspected of glaucoma followed over time from the Duke Glaucoma Registry. METHODS: Linear mixed models were used to investigate the effects of BP on the rates of retinal nerve fiber layer (RNFL) loss from spectral-domain OCT (SD-OCT) over time. Models were adjusted for intraocular pressure (IOP), gender, race, diagnosis, central corneal thickness (CCT), follow-up time, and baseline disease severity. MAIN OUTCOME MEASURE: Effect of mean arterial pressure (MAP), systolic arterial pressure (SAP), and diastolic arterial pressure (DAP) on rates of RNFL loss over time. RESULTS: A total of 157 291 BP visits, 45 408 IOP visits, and 30 238 SD-OCT visits were included. Mean rate of RNFL change was -0.70 µm/year (95% confidence interval, -0.72 to -0.67 µm/year). In univariable models, MAP, SAP, and DAP during follow-up were not significantly associated with rates of RNFL loss. However, when adjusted for mean IOP during follow-up, each 10 mmHg reduction in mean MAP (-0.06 µm/year; P = 0.007) and mean DAP (-0.08 µm/year; P < 0.001) but not SAP (-0.01 µm/year; P = 0.355) was associated with significantly faster rates of RNFL thickness change over time. The effect of the arterial pressure metrics remained significant after additional adjustment for baseline age, diagnosis, sex, race, follow-up time, disease severity, and corneal thickness. CONCLUSIONS: When adjusted for IOP, lower MAP and DAP during follow-up were significantly associated with faster rates of RNFL loss, suggesting that levels of systemic BP may be a significant factor in glaucoma progression.

Structural Changes in the Optic Disc and Macula Detected by Swept-Source Optical Coherence Tomography After Surgical Intraocular Pressure Reduction in Patients with Open-Angle Glaucoma.

PURPOSE: To evaluate the magnitude of change in optic disc, peripapillary retinal nerve fiber layer (RNFL) and macular parameters measured by swept-source optical coherence tomography (SS-OCT) in glaucomatous eyes after filtration surgery, and to determine any possible relationship between these measurements and baseline factors. PATIENTS AND METHODS: This multicenter, prospective, consecutive observational study included patients with open-angle glaucoma who required glaucoma filtering surgery (surgical group, 29 eyes) and those with stable disease (control group, 25 eyes). Patients from the surgical group underwent measurement of optic disc, peripapillary retinal nerve fiber layer (RNFL) and macular parameters before and after surgery. RESULTS: In the surgical group, there was a significant increase in rim area and a significant decrease in the linear cup/disc ratio, vertical cup/disc ratio and cup volume 1 and 2 months postoperatively (p< 0.05). No significant change in the mean RNFL thickness and also sectorial measurements were observed from baseline to 1 and 2 months after surgery (p>0.05). Furthermore, significant increases in macular parameters were observed until 2 months after surgery (p<0.05). No significant changes were observed for all SS-OCT measurements in the control group. There was a significant correlation between the magnitude of the structural measurements change and the IOP reduction for two topographic parameters (rim area and linear cup-disc ratio) and macular average thickness 1 month and 2 months postoperatively. CONCLUSION: In open-angle glaucoma patients submitted to surgical IOP reductions, improvements in topographic and macular OCT parameters measured by SS-OCT were observed for at least 2 months.

RetiNerveNet: using recursive deep learning to estimate pointwise 24-2 visual field data based on retinal structure.

Glaucoma is the leading cause of irreversible blindness in the world, affecting over 70 million people. The cumbersome Standard Automated Perimetry (SAP) test is most frequently used to detect visual loss due to glaucoma. Due to the SAP test's innate difficulty and its high test-retest variability, we propose the RetiNerveNet, a deep convolutional recursive neural network for obtaining estimates of the SAP visual field. RetiNerveNet uses information from the more objective Spectral-Domain Optical Coherence Tomography (SDOCT). RetiNerveNet attempts to trace-back the arcuate convergence of the retinal nerve fibers, starting from the Retinal Nerve Fiber Layer (RNFL) thickness around the optic disc, to estimate individual age-corrected 24-2 SAP values. Recursive passes through the proposed network sequentially yield estimates of the visual locations progressively farther from the optic disc. While all the methods used for our experiments exhibit lower performance for the advanced disease group (possibly due to the "floor effect" for the SDOCT test), the proposed network is observed to be more accurate than all the baselines for estimating the individual visual field values. We further augment the proposed network to additionally predict the SAP Mean Deviation values and also facilitate the assignment of higher weightage to the underrepresented groups in the data. We then study the resulting performance trade-offs of the RetiNerveNet on the early, moderate and severe disease groups.

An objective structural and functional reference standard in glaucoma.

The current lack of consensus for diagnosing glaucoma makes it difficult to develop diagnostic tests derived from deep learning (DL) algorithms. In the present study, we propose an objective definition of glaucomatous optic neuropathy (GON) using clearly defined parameters from optical coherence tomography and standard automated perimetry. We then use the proposed objective definition as reference standard to develop a DL algorithm to detect GON on fundus photos. A DL algorithm was trained to detect GON on fundus photos, using the proposed objective definition as reference standard. The performance was evaluated on an independent test sample with sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and likelihood ratios (LR). The test sample had 2118 fundus photos from 585 eyes of 405 individuals. The AUC to discriminate between GON and normal was 0.92 with sensitivity of 77% at 95% specificity. LRs indicated that the DL algorithm provided large changes in the post-test probability of disease for the majority of eyes. A DL algorithm to evaluate fundus photos had high performance to discriminate GON from normal. The newly proposed objective definition of GON used as reference standard may increase the comparability of diagnostic studies of glaucoma across devices and populations.

Predicting Glaucoma Development With Longitudinal Deep Learning Predictions From Fundus Photographs.

PURPOSE: To assess whether longitudinal changes in a deep learning algorithm's predictions of retinal nerve fiber layer (RNFL) thickness based on fundus photographs can predict future development of glaucomatous visual field defects. DESIGN: Retrospective cohort study. METHODS: This study included 1,072 eyes of 827 glaucoma-suspect patients with an average follow-up of 5.9 ± 3.8 years. All eyes had normal standard automated perimetry (SAP) at baseline. Additional SAP and fundus photographs were acquired throughout follow-up. Conversion to glaucoma was defined as repeatable glaucomatous defects on SAP. An OCT-trained deep learning algorithm (machine to machine, M2M) was used to predict RNFL thicknesses from fundus photographs. Joint longitudinal survival models were used to assess whether baseline and longitudinal change in M2M's RNFL thickness estimates could predict development of visual field defects. RESULTS: A total of 196 eyes (18%) converted to glaucoma during follow-up. The mean rate of change in M2M's predicted RNFL thickness was -1.02 µm/y for converters and -0.67 µm/y for non-converters (P < .001). Baseline and rate of change of predicted RNFL thickness were significantly predictive of conversion to glaucoma, with hazard ratios in the multivariable model of 1.56 per 10 µm lower at baseline (95% CI, 1.33-1.82; P < .001) and 1.99 per 1 µm/y faster loss in thickness during follow-up (95% CI, 1.36-2.93; P < .001). CONCLUSION: Longitudinal changes in a deep learning algorithm's predictions of RNFL thickness measurements based on fundus photographs can be used to predict risk of glaucoma conversion in eyes suspected of having the disease.

Predicting Age From Optical Coherence Tomography Scans With Deep Learning.

Purpose: To assess whether age can be predicted from deep learning analysis of peripapillary spectral-domain optical coherence tomography (SD-OCT) B-scans and to determine the importance of specific retinal areas on the predictions. Methods: Deep learning (DL) convolutional neural networks were developed to predict chronological age in healthy subjects using peripapillary SD-OCT B-scan images. Models were built using the whole B-scan, as well as using specific regions through image ablation. Cross-validation was used for training and testing the model. Mean absolute error (MAE) and correlations between predicted and observed age were used to evaluate model performance. Results: A total of 7271 images from 542 eyes of 278 healthy subjects were included. DL predictions of age using the whole B-scan were strongly correlated with chronological age (MAE = 5.82 years; r = 0.860, P < 0.001). The model also accurately discriminated between the lowest and highest tertiles of age, with an area under the receiver operating characteristic curve of 0.962. In general, class activation maps tended to show a diffuse pattern of activation throughout the scan image. For specific structures of the B-scan, the layers with the strongest correlations with chronological age were the choroid and vitreous (both r = 0.736), whereas retinal nerve fiber layer had the lowest correlation (r = 0.492). Conclusions: A DL algorithm was able to accurately predict age from whole peripapillary SD-OCT B-scans. Translational Relevance: DL models applied to SD-OCT scans suggest that aging appears to affect several layers in the posterior eye segment.

Rates of Glaucomatous Structural and Functional Change From a Large Clinical Population: The Duke Glaucoma Registry Study.

PURPOSE: To investigate rates of structural and functional change in a large clinical population of glaucoma and glaucoma suspect patients. DESIGN: Retrospective cohort. METHODS: Twenty-nine thousand five hundred forty-eight spectral-domain optical coherence tomography (OCT) and 19,812 standard automated perimetry (SAP) tests from 6138 eyes of 3669 patients with ≥6 months of follow-up, 2 good quality spectral-domain OCT peripapillary retinal nerve fiber layer scans, and 2 reliable SAP tests were included. Data were extracted from the Duke Glaucoma Registry, a large database of electronic health records of patients from the Duke Eye Center and satellite clinics. Rates of change for the 2 metrics were obtained using linear mixed models, categorized according to pre-established cutoffs, and analyzed according to the severity of the disease. RESULTS: Average rates of change were -0.73 ± 0.80 µm per year for global retinal nerve fiber layer thickness and -0.09 ± 0.36 dB per year for SAP mean deviation. More than one quarter (26.6%) of eyes were classified as having at least a moderate rate of change by spectral-domain OCT vs 9.1% by SAP (P < .001). In eyes with severe disease, 31.6% were classified as progressing at moderate or faster rates by SAP vs 26.5% by spectral-domain OCT (P = .055). Most eyes classified as fast by spectral-domain OCT were classified as slow by SAP and vice versa. CONCLUSION: Although most patients under routine care had slow rates of progression, a substantial proportion had rates that could potentially result in major losses if sustained over time. Both structural and functional tests should be used to monitor glaucoma, and spectral-domain OCT still has a relevant role in detecting fast progressors in advanced disease.

Impact of Intraocular Pressure Control on Rates of Retinal Nerve Fiber Layer Loss in a Large Clinical Population.

PURPOSE: To investigate the impact of intraocular pressure (IOP) control on rates of change of spectral-domain OCT (SD-OCT) retinal nerve fiber layer (RNFL) thickness in a large clinical population. DESIGN: Retrospective cohort study. PARTICIPANTS: A total of 85 835 IOP measurements and 60 223 SD-OCT tests from 14 790 eyes of 7844 patients. METHODS: Data were extracted from the Duke Glaucoma Registry, a large database of electronic medical records of patients with glaucoma and suspected disease followed over time at the Duke Eye Center and satellite clinics. All records from patients with a minimum of 6 months of follow-up and at least 2 good-quality SD-OCT scans and 2 clinical visits with Goldmann applanation tonometry were included. Eyes were categorized according to the frequency of visits with IOP below cutoffs of 21 mmHg, 18 mmHg, and 15 mmHg over time. Rates of change for global RNFL thickness were obtained using linear mixed models and classified as slow if change was slower than -1.0 µm/year; moderate if between -1.0 and -2.0 µm/year; and fast if faster than -2.0 µm/year. Multivariable models were adjusted for age, gender, race, diagnosis, central corneal thickness, follow-up time, and baseline disease severity. MAIN OUTCOME MEASURES: Rates of change in SD-OCT RNFL thickness according to levels of IOP control. RESULTS: Eyes had a mean follow-up of 3.5±1.9 years. Average rate of change in RNFL thickness was -0.68±0.59 µm/year. Each 1 mmHg higher mean IOP was associated with 0.05 µm/year faster RNFL loss (P < 0.001) after adjustment for potentially confounding variables. For eyes that had fast progression, 41% of them had IOP <21 mmHg in all visits during follow-up, whereas 20% of them had all visits with IOP <18 mmHg, but only 9% of them had all visits with IOP <15 mmHg. CONCLUSIONS: Intraocular pressure was significantly associated with rates of progressive RNFL loss in a large clinical population. Eyes with stricter IOP control over follow-up visits had a smaller chance of exhibiting fast deterioration. Our findings may assist clinicians in establishing target pressures in clinical practice.

Detection of Progressive Glaucomatous Optic Nerve Damage on Fundus Photographs with Deep Learning.

PURPOSE: To investigate whether predictions of retinal nerve fiber layer (RNFL) thickness obtained from a deep learning model applied to fundus photographs can detect progressive glaucomatous changes over time. DESIGN: Retrospective cohort study. PARTICIPANTS: Eighty-six thousand one hundred twenty-three pairs of color fundus photographs and spectral-domain (SD) OCT images collected during 21 232 visits from 8831 eyes of 5529 patients with glaucoma or glaucoma suspects. METHODS: A deep learning convolutional neural network was trained to assess fundus photographs and to predict SD OCT global RNFL thickness measurements. The model then was tested on an independent sample of eyes that had longitudinal follow-up with both fundus photography and SD OCT. The ability to detect eyes that had statistically significant slopes of SD OCT change was assessed by receiver operating characteristic (ROC) curves. The repeatability of RNFL thickness predictions was investigated by measurements obtained from multiple photographs that had been acquired during the same day. MAIN OUTCOME MEASURES: The relationship between change in predicted RNFL thickness from photographs and change in SD OCT RNFL thickness over time. RESULTS: The test sample consisted of 33 466 pairs of fundus photographs and SD OCT images collected during 7125 visits from 1147 eyes of 717 patients. Eyes in the test sample were followed up for an average of 5.3 ± 3.3 years, with an average of 6.2 ± 3.8 visits. A significant correlation was found between change over time in predicted and observed RNFL thickness (r = 0.76; 95% confidence interval [CI], 0.70-0.80; P < 0.001). Retinal nerve fiber layer predictions showed an ROC curve area of 0.86 (95% CI, 0.83-0.88) to discriminate progressors from nonprogressors. For detecting fast progressors (slope faster than 2 µm/year), the ROC curve area was 0.96 (95% CI, 0.94-0.98), with a sensitivity of 97% for 80% specificity and 85% for 90% specificity. For photographs obtained at the same visit, the intraclass correlation coefficient was 0.946 (95% CI, 0.940-0.952), with a coefficient of variation of 3.2% (95% CI, 3.1%-3.3%). CONCLUSIONS: A deep learning model was able to obtain objective and quantitative estimates of RNFL thickness that correlated well with SD OCT measurements and potentially could be used to monitor for glaucomatous changes over time.

Artificial Intelligence Mapping of Structure to Function in Glaucoma.

Purpose: To develop an artificial intelligence (AI)-based structure-function (SF) map relating retinal nerve fiber layer (RNFL) damage on spectral domain optical coherence tomography (SDOCT) to functional loss on standard automated perimetry (SAP). Methods: The study included 26,499 pairs of SAP and SDOCT from 15,173 eyes of 8878 patients with glaucoma or suspected of having the disease extracted from the Duke Glaucoma Registry. The data set was randomly divided at the patient level in training and test sets. A convolutional neural network (CNN) was initially trained and validated to predict the 52 sensitivity threshold points of the 24-2 SAP from the 768 RNFL thickness points of the SDOCT peripapillary scan. Simulated localized RNFL defects of varied locations and depths were created by modifying the normal average peripapillary RNFL profile. The simulated profiles were then fed to the previously trained CNN, and the topographic SF relationships between structural defects and SAP functional losses were investigated. Results: The CNN predictions had an average correlation coefficient of 0.60 (P < 0.001) with the measured values from SAP and a mean absolute error of 4.25 dB. Simulated RNFL defects led to well-defined arcuate or paracentral visual field losses in the opposite hemifield, which varied according to the location and depth of the simulations. Conclusions: A CNN was capable of predicting SAP sensitivity thresholds from SDOCT RNFL thickness measurements and generate an SF map from simulated defects. Translational Relevance: AI-based SF map improves the understanding of how SDOCT losses translate into detectable SAP damage.

The Relationship Between Asymmetries of Corneal Properties and Rates of Visual Field Progression in Glaucoma Patients.

PRéCIS:: In this study, asymmetries in corneal hysteresis (CH) between eyes of glaucoma patients were significantly associated with asymmetries in rates of visual field loss, suggesting a role of hysteresis as a risk factor for disease progression. PURPOSE: The purpose of this study was to investigate the relationship between asymmetries in rates of glaucoma progression and asymmetries of corneal properties between eyes of subjects with primary open-angle glaucoma. PARTICIPANTS AND METHODS: This prospective study followed 126 binocular subjects with glaucoma for an average of 4.3±0.8 years. CH was measured at baseline using the Ocular Response Analyzer. Standard automated perimetry (SAP) and intraocular pressure were measured at baseline and every 6 months. Rates of visual field progression were calculated using ordinary least square regression of SAP mean deviation (MD) values over time for each eye. Eyes were defined as "better" and "worse" based on the slopes of SAP MD. Pearson correlation test, and univariable and multivariable regression models were used to investigate the relationship between inter-eye asymmetry in CH and central corneal thickness and inter-eye differences in rates of visual field progression. RESULTS: Only asymmetry of CH was significantly associated with the asymmetry in SAP MD rates of change between eyes (r=0.22; P=0.01). In a multivariable model adjusting for age, race, central corneal thickness, mean intraocular pressure and baseline disease severity, CH asymmetry remained significantly associated with asymmetric progression (P=0.032). CONCLUSION: CH asymmetry between eyes was associated with asymmetry on rates of visual field change, providing further support for the role of CH as a risk factor for glaucoma progression.

Comparing the Rule of 5 to Trend-based Analysis for Detecting Glaucoma Progression on OCT.

PURPOSE: The rule of 5 is a simple rule for detecting retinal nerve fiber layer (RNFL) change on spectral-domain OCT (SD-OCT), in which a loss of 5 µm of global RNFL on a follow-up test is considered evidence of significant change when compared with the baseline. The rule is based on short-term test-retest variability of SD-OCT and is often used in clinical practice. The purpose of this study was to compare the rule of 5 with trend-based analysis of global RNFL thickness over time for detecting glaucomatous progression. DESIGN: Prospective cohort. PARTICIPANTS: A total of 300 eyes of 210 glaucoma subjects followed for an average of 5.4±1.5 years with a median of 11 (interquartile range, 7-14) visits. METHODS: Trend-based analysis was performed by ordinary least-squares (OLS) linear regression of global RNFL thickness over time. For estimation of specificity, false-positives were obtained by assessing for progression on series of randomly permutated follow-up visits for each eye, which removes any systematic trend over time. The specificity of trend-based analysis was matched to that of the rule of 5 to allow meaningful comparison of the "hit rate," or the proportion of glaucoma eyes categorized as progressing at each time point, using the original sequence of visits. MAIN OUTCOME MEASURES: Comparison between hit rates of trend-analysis versus rule of 5 at matched specificity. RESULTS: After 5 years, the simple rule of 5 identified 37.5% of eyes as progressing at a specificity of 81.1%. At the same specificity, the hit rate for trend-based analysis was significantly greater than that of the rule of 5 (62.9% vs. 37.5%; P < 0.001). If the rule of 5 was required to be repeatable on a consecutive test, specificity improved to 93.4%, but hit rate decreased to 21.0%. At this higher specificity, trend-based analysis still had a significantly greater hit rate than the rule of 5 (47.4% vs. 21.0%, respectively; P < 0.001). CONCLUSIONS: Trend-based analysis was superior to the simple rule of 5 for identifying progression in glaucoma eyes and should be preferred as a method for longitudinal assessment of global SD-OCT RNFL change over time.

Assessment of a Segmentation-Free Deep Learning Algorithm for Diagnosing Glaucoma From Optical Coherence Tomography Scans.

Importance: Conventional segmentation of the retinal nerve fiber layer (RNFL) is prone to errors that may affect the accuracy of spectral-domain optical coherence tomography (SD-OCT) scans in detecting glaucomatous damage. Objective: To develop a segmentation-free deep learning (DL) algorithm for assessment of glaucomatous damage using the entire circle B-scan image from SD-OCT. Design, Setting, and Participants: This cross-sectional study at a single institution used data from SD-OCT images of eyes with glaucoma (perimetric and preperimetric) and normal eyes. The data set was randomly split at the patient level into a training (50%), validation (20%), and test data set (30%). Data were collected from March 2008 to April 2019, and analysis began April 2018. Exposures: A convolutional neural network was trained to discriminate glaucomatous from normal eyes using the SD-OCT circle B-scan without segmentation lines. Main Outcomes and Measures: The ability to discriminate glaucoma from healthy eyes was evaluated by comparing the area under the receiver operating characteristic curve and sensitivity at 80% or 95% specificity for the DL algorithm's predicted probability of glaucoma vs conventional RNFL thickness parameters given by SD-OCT software. The performance was also assessed in preperimetric glaucoma, as well as by visual field severity using Hodapp-Parrish-Anderson criteria. Results: A total of 20 806 SD-OCT images from 1154 eyes of 635 individuals (612 [53%] with glaucoma and 542 normal eyes [47%]) were included. The mean (SD) age at SD-OCT scan was 70.8 (10.4) years in individuals with glaucoma and 55.8 (14.1) years in controls. There were 187 women (53.3%) in the glaucoma group and 165 (59.8%) in the control group. Of 612 eyes with glaucoma, 432 (70.4%) had perimetric and 180 (29.6%) had preperimetric glaucoma. The DL algorithm had a significantly higher area under the receiver operating characteristic curve than global RNFL thickness (0.96 vs 0.87; difference = 0.08 [95% CI, 0.04-0.12]) and each RNFL thickness sector for discriminating between glaucoma and controls (all P < .001). At 95% specificity, the DL algorithm (81%; 95% CI, 64%-97%) was more sensitive than global RNFL thickness (67%; 95% CI, 58%-76%). The areas under the receiver operating characteristic curve were also significantly greater for the DL algorithm compared with RNFL thickness at each stage of disease, especially preperimetric and mild perimetric glaucoma. Conclusions and Relevance: A segmentation-free DL algorithm performed better than conventional RNFL thickness parameters for diagnosing glaucomatous damage on OCT scans, especially in early disease. Future studies should investigate how such an approach contributes to diagnostic decisions when combined with other relevant clinical information, such as risk factors and perimetry results.

Quantification of Retinal Nerve Fibre Layer Thickness on Optical Coherence Tomography with a Deep Learning Segmentation-Free Approach.

This study describes a segmentation-free deep learning (DL) algorithm for measuring retinal nerve fibre layer (RNFL) thickness on spectral-domain optical coherence tomography (SDOCT). The study included 25,285 B-scans from 1,338 eyes of 706 subjects. Training was done to predict RNFL thickness from raw unsegmented scans using conventional RNFL thickness measurements from good quality images as targets, forcing the DL algorithm to learn its own representation of RNFL. The algorithm was tested in three different sets: (1) images without segmentation errors or artefacts, (2) low-quality images with segmentation errors, and (3) images with other artefacts. In test set 1, segmentation-free RNFL predictions were highly correlated with conventional RNFL thickness (r = 0.983, P < 0.001). In test set 2, segmentation-free predictions had higher correlation with the best available estimate (tests with good quality taken in the same date) compared to those from the conventional algorithm (r = 0.972 vs. r = 0.829, respectively; P < 0.001). Segmentation-free predictions were also better in test set 3 (r = 0.940 vs. r = 0.640, P < 0.001). In conclusion, a novel segmentation-free algorithm to extract RNFL thickness performed similarly to the conventional method in good quality images and better in images with errors or other artefacts.

Comparison of Short- And Long-Term Variability in Standard Perimetry and Spectral Domain Optical Coherence Tomography in Glaucoma.

PURPOSE: To assess short- and long-term variability on standard automated perimetry (SAP) and spectral domain optical coherence tomography (SD-OCT) in glaucoma. DESIGN: Prospective cohort. METHODS: Ordinary least squares linear regression of SAP mean deviation (MD) and SD-OCT global retinal nerve fiber layer (RNFL) thickness were fitted over time for sequential tests conducted within 5 weeks (short-term testing) and annually (long-term testing). Residuals were obtained by subtracting the predicted and observed values, and each patient's standard deviation (SD) of the residuals was used as a measure of variability. Wilcoxon signed-rank test was performed to test the hypothesis of equality between short- and long-term variability. RESULTS: A total of 43 eyes of 43 glaucoma subjects were included. Subjects had a mean 4.5 ± 0.8 SAP and OCT tests for short-term variability assessment. For long-term variability, the same number of tests were performed and results annually collected over an average of 4.0 ± 0.8 years. The average SD of the residuals was significantly higher in the long-term than in the short-term period for both tests: 1.05 ± 0.70 dB vs. 0.61 ± 0.34 dB, respectively (P < 0.001) for SAP MD and 1.95 ± 1.86 µm vs. 0.81 ± 0.56 µm, respectively (P < 0.001) for SD-OCT RNFL thickness. CONCLUSIONS: Long-term variability was higher than short-term variability on SD-OCT and SAP. Because current event-based algorithms for detection of glaucoma progression on SAP and SD-OCT have relied on short-term variability data to establish their normative databases, these algorithms may be underestimating the variability in the long-term and thus may overestimate progression over time.

Human Versus Machine: Comparing a Deep Learning Algorithm to Human Gradings for Detecting Glaucoma on Fundus Photographs.

PURPOSE: To compare the diagnostic performance of human gradings vs predictions provided by a machine-to-machine (M2M) deep learning (DL) algorithm trained to quantify retinal nerve fiber layer (RNFL) damage on fundus photographs. DESIGN: Evaluation of a machine learning algorithm. METHODS: An M2M DL algorithm trained with RNFL thickness parameters from spectral-domain optical coherence tomography was applied to a subset of 490 fundus photos of 490 eyes of 370 subjects graded by 2 glaucoma specialists for the probability of glaucomatous optical neuropathy (GON), and estimates of cup-to-disc (C/D) ratios. Spearman correlations with standard automated perimetry (SAP) global indices were compared between the human gradings vs the M2M DL-predicted RNFL thickness values. The area under the receiver operating characteristic curves (AUC) and partial AUC for the region of clinically meaningful specificity (85%-100%) were used to compare the ability of each output to discriminate eyes with repeatable glaucomatous SAP defects vs eyes with normal fields. RESULTS: The M2M DL-predicted RNFL thickness had a significantly stronger absolute correlation with SAP mean deviation (rho=0.54) than the probability of GON given by human graders (rho=0.48; P < .001). The partial AUC for the M2M DL algorithm was significantly higher than that for the probability of GON by human graders (partial AUC = 0.529 vs 0.411, respectively; P = .016). CONCLUSION: An M2M DL algorithm performed as well as, if not better than, human graders at detecting eyes with repeatable glaucomatous visual field loss. This DL algorithm could potentially replace human graders in population screening efforts for glaucoma.

Detecting Retinal Nerve Fibre Layer Segmentation Errors on Spectral Domain-Optical Coherence Tomography with a Deep Learning Algorithm.

In this study we developed a deep learning (DL) algorithm that detects errors in retinal never fibre layer (RNFL) segmentation on spectral-domain optical coherence tomography (SDOCT) B-scans using human grades as the reference standard. A dataset of 25,250 SDOCT B-scans reviewed for segmentation errors by human graders was randomly divided into validation plus training (50%) and test (50%) sets. The performance of the DL algorithm was evaluated in the test sample by outputting a probability of having a segmentation error for each B-scan. The ability of the algorithm to detect segmentation errors was evaluated with the area under the receiver operating characteristic (ROC) curve. Mean DL probabilities of segmentation error in the test sample were 0.90 ± 0.17 vs. 0.12 ± 0.22 (P < 0.001) for scans with and without segmentation errors, respectively. The DL algorithm had an area under the ROC curve of 0.979 (95% CI: 0.974 to 0.984) and an overall accuracy of 92.4%. For the B-scans with severe segmentation errors in the test sample, the DL algorithm was 98.9% sensitive. This algorithm can help clinicians and researchers review images for artifacts in SDOCT tests in a timely manner and avoid inaccurate diagnostic interpretations.