Developing a privacy-preserving deep learning model for glaucoma detection: a multicentre study with federated learning.

BACKGROUND: Deep learning (DL) is promising to detect glaucoma. However, patients' privacy and data security are major concerns when pooling all data for model development. We developed a privacy-preserving DL model using the federated learning (FL) paradigm to detect glaucoma from optical coherence tomography (OCT) images. METHODS: This is a multicentre study. The FL paradigm consisted of a 'central server' and seven eye centres in Hong Kong, the USA and Singapore. Each centre first trained a model locally with its own OCT optic disc volumetric dataset and then uploaded its model parameters to the central server. The central server used FedProx algorithm to aggregate all centres' model parameters. Subsequently, the aggregated parameters are redistributed to each centre for its local model optimisation. We experimented with three three-dimensional (3D) networks to evaluate the stabilities of the FL paradigm. Lastly, we tested the FL model on two prospectively collected unseen datasets. RESULTS: We used 9326 volumetric OCT scans from 2785 subjects. The FL model performed consistently well with different networks in 7 centres (accuracies 78.3%-98.5%, 75.9%-97.0%, and 78.3%-97.5%, respectively) and stably in the 2 unseen datasets (accuracies 84.8%-87.7%, 81.3%-84.8%, and 86.0%-87.8%, respectively). The FL model achieved non-inferior performance in classifying glaucoma compared with the traditional model and significantly outperformed the individual models. CONCLUSION: The 3D FL model could leverage all the datasets and achieve generalisable performance, without data exchange across centres. This study demonstrated an OCT-based FL paradigm for glaucoma identification with ensured patient privacy and data security, charting another course toward the real-world transition of artificial intelligence in ophthalmology.

Pediatric Glaucoma-From Screening, Early Detection to Management.

Pediatric glaucoma (PG) covers a rare and heterogeneous group of diseases with variable causes and presentations. Delayed diagnosis of PG could lead to blindness, bringing emotional and psychological burdens to patients' caregivers. Recent genetic studies identified novel causative genes, which may provide new insight into the etiology of PG. More effective screening strategies could be beneficial for timely diagnosis and treatment. New findings on clinical characteristics and the latest examination instruments have provided additional evidence for diagnosing PG. In addition to IOP-lowering therapy, managing concomitant amblyopia and other associated ocular pathologies is essential to achieve a better visual outcome. Surgical treatment is usually required although medication is often used before surgery. These include angle surgeries, filtering surgeries, minimally invasive glaucoma surgeries, cyclophotocoagulation, and deep sclerectomy. Several advanced surgical therapies have been developed to increase success rates and decrease postoperative complications. Here, we review the classification and diagnosis, etiology, screening, clinical characteristics, examinations, and management of PG.

Clinical outcomes in children and adolescents referred for increased cup:disk ratio at a tertiary referral center.

PURPOSE: To report clinical outcomes and risk factors for glaucoma in children and adolescents referred for increased cup:disk ratios (CDRs) to a tertiary referral center. METHODS: This retrospective, single-center study examined all pediatric patients evaluated for increased CDR at Wills Eye Hospital. Patients who had previous known ocular disease were excluded. Demographic data, including sex, age, and race/ethnicity were recorded, as were baseline and follow-up ophthalmic examination findings, including intraocular pressure (IOP), CDR, diurnal curve, gonioscopy findings, and refractive error. Risks of glaucoma diagnosis based on these data were analyzed. RESULTS: A total of 167 patients were included, of whom 6 were found to have glaucoma. Despite more than 2 years' follow-up on 61 patients, all glaucoma patients were identified within the first 3 months of evaluation. Baseline IOP was statistically significantly higher in glaucomatous patients than nonglaucomatous patients (28 ± 7 vs 15 ± 4, resp. [P = 0.0002]), as was maximum IOP on diurnal curve (24 ± 3 vs 17 ± 3 [P = 0.0005]). CONCLUSIONS: In our study cohort, diagnosis of glaucoma was apparent in the first year of evaluation. Baseline IOP and maximal IOP on diurnal curve were found to be statistically significantly associated with glaucoma diagnosis in pediatric patients referred for increased CDR.

Risk of Normal Tension Glaucoma Progression From Automated Baseline Retinal-Vessel Caliber Analysis: A Prospective Cohort Study.

PURPOSE: To determine the relationship between baseline retinal-vessel calibers computed by a deep-learning system and the risk of normal tension glaucoma (NTG) progression. DESIGN: Prospective cohort study. METHODS: Three hundred and ninety eyes from 197 patients with NTG were followed up for at least 24 months. Retinal-vessel calibers (central retinal arteriolar equivalent [CRAE] and central retinal venular equivalent [CRVE]) were computed from fundus photographs at baseline using a previously validated deep-learning system. Retinal nerve fiber layer (RNFL) thickness and visual field (VF) were evaluated semiannually. The Cox proportional-hazards model was used to evaluate the relationship of baseline retinal-vessel calibers to the risk of glaucoma progression. RESULTS: Over a mean follow-up period of 34.36 ± 5.88 months, 69 NTG eyes (17.69%) developed progressive RNFL thinning and 22 eyes (5.64%) developed VF deterioration. In the multivariable Cox regression analysis adjusting for age, gender, intraocular pressure, mean ocular perfusion pressure, systolic blood pressure, axial length, standard automated perimetry mean deviation, and RNFL thickness, narrower baseline CRAE (hazard ratio per SD decrease [95% confidence interval], 1.36 [1.01-1.82]) and CRVE (1.35 [1.01-1.80]) were associated with progressive RNFL thinning and narrower baseline CRAE (1.98 [1.17-3.35]) was associated with VF deterioration. CONCLUSION: In this study, each SD decrease in the baseline CRAE or CRVE was associated with a more than 30% increase in the risk of progressive RNFL thinning and a more than 90% increase in the risk of VF deterioration during the follow-up period. Baseline attenuation of retinal vasculature in NTG eyes was associated with subsequent glaucoma progression. High-throughput deep-learning-based retinal vasculature analysis demonstrated its clinical utility for NTG risk assessment.

Three-Dimensional Multi-Task Deep Learning Model to Detect Glaucomatous Optic Neuropathy and Myopic Features From Optical Coherence Tomography Scans: A Retrospective Multi-Centre Study.

Purpose: We aim to develop a multi-task three-dimensional (3D) deep learning (DL) model to detect glaucomatous optic neuropathy (GON) and myopic features (MF) simultaneously from spectral-domain optical coherence tomography (SDOCT) volumetric scans. Methods: Each volumetric scan was labelled as GON according to the criteria of retinal nerve fibre layer (RNFL) thinning, with a structural defect that correlated in position with the visual field defect (i.e., reference standard). MF were graded by the SDOCT en face images, defined as presence of peripapillary atrophy (PPA), optic disc tilting, or fundus tessellation. The multi-task DL model was developed by ResNet with output of Yes/No GON and Yes/No MF. SDOCT scans were collected in a tertiary eye hospital (Hong Kong SAR, China) for training (80%), tuning (10%), and internal validation (10%). External testing was performed on five independent datasets from eye centres in Hong Kong, the United States, and Singapore, respectively. For GON detection, we compared the model to the average RNFL thickness measurement generated from the SDOCT device. To investigate whether MF can affect the model's performance on GON detection, we conducted subgroup analyses in groups stratified by Yes/No MF. The area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, and accuracy were reported. Results: A total of 8,151 SDOCT volumetric scans from 3,609 eyes were collected. For detecting GON, in the internal validation, the proposed 3D model had significantly higher AUROC (0.949 vs. 0.913, p < 0.001) than average RNFL thickness in discriminating GON from normal. In the external testing, the two approaches had comparable performance. In the subgroup analysis, the multi-task DL model performed significantly better in the group of "no MF" (0.883 vs. 0.965, p-value < 0.001) in one external testing dataset, but no significant difference in internal validation and other external testing datasets. The multi-task DL model's performance to detect MF was also generalizable in all datasets, with the AUROC values ranging from 0.855 to 0.896. Conclusion: The proposed multi-task 3D DL model demonstrated high generalizability in all the datasets and the presence of MF did not affect the accuracy of GON detection generally.

Characterization of macular choroid in normal-tension glaucoma: a swept-source optical coherence tomography study.

PURPOSE: To investigate the features of the small-to-medium (choriocapillaris and Sattler's layer) and large (Haller's layer) sized vessel layers of the macular choroid in normal-tension glaucoma (NTG) patients using swept-source optical coherence tomography (SS-OCT). METHODS: We conducted an observational cross-sectional study in 234 NTG eyes from 134 patients, and 203 normal eyes from 109 non-glaucomatous control subjects. We used Image J to segment the choroidal layer and then a Python script to measure the average macular choroidal thickness (MCT) and choroidal vascular index (CVI) of two vessel layers at five different locations. Quantitative optical coherence tomography angiography (OCTA) metrics for the superficial capillary plexus (SCP) and deep capillary plexus (DCP) at macular region were also measured by a customized MATLAB program. Generalized estimating equations (GEE) models were performed to determine ocular and demographic factors associated with the choroidal metrics, adjusting for inter-eye correlation. RESULTS: Significant average MCT thinning was found in NTG eyes at all five locations of the two layers, in comparison with controls (all p ≤ 0.05). In addition, compared with control eyes, significant decrease in CVI was found in NTG eyes at all five sections of the large sized vessel layer: 500 µm nasal and temporal to macula (p = 0.002), 1500 µm nasal (p < 0.001), 2500 µm nasal (p = 0.001), 1500 µm temporal (p < 0.001) and 2500 µm temporal (p = 0.004). In contrast, no significant CVI difference was detected in the small-to-medium sized vessel layer. In the comparison of OCTA metrics of SCP and DCP at macular region between NTG and normal eyes, there were no significant difference of parafoveal vessel density (VD), foveal avascular zone (FAZ) area, FAZ circularity and fractal dimension (FD) in both layers. CONCLUSION: We found significant alterations in macular choroidal vascularity (reduced CVI and thinner layer) in NTG patients. Such alterations are more pronounced in the Haller's layer, rather than the choriocapillaris & Sattler's layer, in NTG. Choroidal layer may be more related to vasculature changes at macular region in NTG.

Efficacy and safety of oral valganciclovir in cytomegalovirus anterior uveitis with uncontrolled intraocular pressure.

BACKGROUND/AIMS: While cytomegalovirus (CMV) anterior uveitis (AU) patients often require glaucoma surgery, the effectiveness of systemic anti-viral in long-term intraocular pressure (IOP) control is not well established. Our study aims to identify the 2-year efficacy and safety of oral valganciclovir in CMV AU with uncontrolled IOP. METHODS: In this retrospective case series, one eye from each of 17 immunocompetent PCR-proven patients with CMV AU who received a single course of oral valganciclovir for 20-148 days for medically uncontrolled IOP during 2008-2018 were identified. They were examined at baseline, week 2, months 1, 2 and 3, then every 3 months up to 2 years after commencement of valganciclovir, or until IOP-lowering procedure. RESULTS: Median baseline IOP and IOP-lowering medication were 27.0 mm Hg (IQR: 22.9-31.0 mm Hg), and 4.0, respectively. IOP was significantly lower than baseline from 2 weeks to 12 months and at 21 and 24 months after starting valganciclovir (p=0.001 to 0.041, Wilcoxon sign-rank test), with 16.9-46.0% median IOP reduction. Seven (41.2%) and six (35.3%) patients had IOP≤21 mm Hg with same, or reduced, topical medications by 12 and 24 months, respectively. Median time to IOP-lowering intervention or second course of valganciclovir was 12.4 months. There was no serious medication-related adverse event. Common side effects included reduced monocyte count (9 patients) and deranged renal function/electrolytes (5 patients). IOP spike and wound leak occurred in 35.5% and 29.4% of patients, respectively, after diagnostic aqueous tap. CONCLUSION: In CMV AU with uncontrolled IOP, >1/3 of the patients avoided glaucoma surgery over 2 years with a course of oral valganciclovir.

Global assessment of arteriolar, venular and capillary changes in normal tension glaucoma.

Microcirculatory insufficiency has been hypothesized in glaucoma pathogenesis. There is a scarcity of data to comprehensively examine the changes in retinal microvasculature and its role in normal tension glaucoma (NTG). We conducted a cross-sectional case-control study and included 168 eyes from 100 NTG patients and 68 healthy subjects. Quantitative retinal arteriolar and venular metrics were measured from retinal photographs using a computer-assisted program. Radial peripapillary capillary network was imaged with OCT-A and quantitative capillary metrics (circumpapillary vessel density (cpVD) and circumpapillary fractal dimension (cpFD)) were measured with a customized MATLAB program. We found that NTG was associated with decreased arteriolar and venular tortuosity, arteriolar branching angle, cpVD and cpFD. Decreased venular caliber, arteriolar and venular branching angles, cpVD and cpFD were associated with thinner average RNFL thickness. Decreased arteriolar and venular branching angles, cpVD and cpFD were also associated with worse standard automated perimetry measurements (mean deviation and visual field index). Compared with retinal arteriolar and venular metrics, regression models based on OCT-A capillary metrics consistently showed stronger associations with NTG and structural and functional measurements in NTG. We concluded that NTG eyes showed generalized microvascular attenuations, in which OCT-A capillary metrics attenuations were more prominent and strongly associated with NTG.

Detection of glaucomatous optic neuropathy with spectral-domain optical coherence tomography: a retrospective training and validation deep-learning analysis.

BACKGROUND: Spectral-domain optical coherence tomography (SDOCT) can be used to detect glaucomatous optic neuropathy, but human expertise in interpretation of SDOCT is limited. We aimed to develop and validate a three-dimensional (3D) deep-learning system using SDOCT volumes to detect glaucomatous optic neuropathy. METHODS: We retrospectively collected a dataset including 4877 SDOCT volumes of optic disc cube for training (60%), testing (20%), and primary validation (20%) from electronic medical and research records at the Chinese University of Hong Kong Eye Centre (Hong Kong, China) and the Hong Kong Eye Hospital (Hong Kong, China). Residual network was used to build the 3D deep-learning system. Three independent datasets (two from Hong Kong and one from Stanford, CA, USA), including 546, 267, and 1231 SDOCT volumes, respectively, were used for external validation of the deep-learning system. Volumes were labelled as having or not having glaucomatous optic neuropathy according to the criteria of retinal nerve fibre layer thinning on reliable SDOCT images with position-correlated visual field defect. Heatmaps were generated for qualitative assessments. FINDINGS: 6921 SDOCT volumes from 1 384 200 two-dimensional cross-sectional scans were studied. The 3D deep-learning system had an area under the receiver operation characteristics curve (AUROC) of 0·969 (95% CI 0·960-0·976), sensitivity of 89% (95% CI 83-93), specificity of 96% (92-99), and accuracy of 91% (89-93) in the primary validation, outperforming a two-dimensional deep-learning system that was trained on en face fundus images (AUROC 0·921 [0·905-0·937]; p<0·0001). The 3D deep-learning system performed similarly in the external validation datasets, with AUROCs of 0·893-0·897, sensitivities of 78-90%, specificities of 79-86%, and accuracies of 80-86%. The heatmaps of glaucomatous optic neuropathy showed that the learned features by the 3D deep-learning system used for detection of glaucomatous optic neuropathy were similar to those used by clinicians. INTERPRETATION: The proposed 3D deep-learning system performed well in detection of glaucomatous optic neuropathy in both primary and external validations. Further prospective studies are needed to estimate the incremental cost-effectiveness of incorporation of an artificial intelligence-based model for glaucoma screening. FUNDING: Hong Kong Research Grants Council.

Correlation of Intraocular Pressure Between Both Eyes After Bilateral Selective Laser Trabeculoplasty in Open-angle Glaucoma.

PURPOSE: To investigate the correlation of intraocular pressure (IOP)-lowering effects of selective laser trabeculoplasty (SLT) between the 2 eyes treated with SLT in open-angle glaucoma (OAG). METHODS: This prospective cohort study sequentially recruited subjects with bilateral OAG. All subjects received a single session of 360-degree SLT treatment. Success was defined as IOP reduction of ≥20%. Spearman correlation was used to compare the following parameters between the 2 eyes following SLT: IOP at day 1, 1 week, and 1 month; percentage of success; and IOP reduction. RESULTS: In 84 eyes of 42 subjects that received bilateral SLT treatment, both eyes had statistically comparable baseline characteristics. There were significant correlations between the IOP in both eyes at all time intervals following SLT as well as for the percentage of IOP reduction and the success rate at 1 month after SLT (all r≥0.7, P<0.0001). A total of 42.9% of subjects had bilateral success and 38.1% had bilateral nonsuccess with a significant correlation between both eyes in these 2 groups (Spearman r>0.6, P<0.02). Nineteen percent had success in 1 eye and nonsuccess in the fellow eye with an inverse correlation between the 2 eyes (Spearman r=-0.7, P=0.03). CONCLUSION: There is a strong and significant correlation in the IOP-lowering response to SLT between both eyes in near 80% of treated OAG subjects, whereas near 20% had an asymmetrical and inverse response to SLT between both the eyes.

Optimal selective laser trabeculoplasty energy for maximal intraocular pressure reduction in open-angle glaucoma.

PURPOSE: To identify the optimal energy level to be used in selective laser trabeculoplasty (SLT) for maximal intraocular pressure (IOP) reduction in open-angle glaucoma (OAG) patients. PATIENTS AND METHODS: This cohort sequentially recruited OAG subjects in Hong Kong, China during 2011 to 2012. All subjects received a single session of SLT with near confluent spots to 360 degrees of the trabecular meshwork. An initial energy of 0.8 mJ was titrated until bubble formation was just visible. The main outcomes included: change in IOP (pre-SLT to 1 mo post-SLT) and total SLT energy (SLT spots multiplied by the mean energy). For statistical analysis, only the right eye of each subject was used. Bandwidth selection by generalized cross-validation was used to determine the optimal interval and point of total SLT energy that resulted in the largest IOP reduction. RESULTS: A total of 49 Chinese OAG subjects had a mean age of 64.2±11.1 years. The pre-SLT IOP was 17.1±2.9 mm Hg while on 1.9±1.1 types of antiglaucoma eye drops. The mean total energy was 167.1±41.4 mJ (171.5±41.2 spots at 1.0±0.06 mJ). The 1 month post-SLT IOP was 13.5±2.8 mm Hg. The percentage of SLT success was 57.1% (28/49). The 95% confidence band by bootstrap method was plotted showing that a total energy between 214.6 and 234.9 mJ significantly decreased the IOP>25%, with the optimal total energy at 226.1 mJ. CONCLUSIONS: A higher SLT energy, in the range of 214.6 to 234.9 mJ, seems to be associated with an improved IOP-lowering response. Further randomized control trials with treatment stratification are needed to confirm these results.