Search

Purpose: The Ocular Hypertension Treatment Study (OHTS) identified risk factors for primary open-angle glaucoma (POAG) in patients with ocular hypertension, including pattern standard deviation (PSD). Archetypal analysis, an unsupervised machine learning method, may offer a more interpretable approach to risk stratification by identifying patterns in baseline visual fields (VFs)., Methods: There were 3272 eyes available in the OHTS. Archetypal analysis was applied using 24-2 baseline VFs, and model selection was performed with cross-validation. Decomposition coefficients for archetypes (ATs) were calculated. A penalized Cox proportional hazards model was implemented to select discriminative ATs. The AT model was compared to the OHTS model. Associations were identified between ATs with both POAG onset and VF progression, defined by mean deviation change per year., Results: We selected 8494 baseline VFs. Optimal AT count was 19. The highest prevalence ATs were AT9, AT11, and AT7. The AT-based prediction model had a C-index of 0.75 for POAG onset. Multivariable models demonstrated that a one-interquartile range increase in the AT5 (hazard ratio [HR] = 1.14; 95% confidence interval [CI], 1.04-1.25), AT8 (HR = 1.22; 95% CI, 1.09-1.37), AT15 (HR = 1.26; 95% CI, 1.12-1.41), and AT17 (HR = 1.17; 95% CI, 1.03-1.31) coefficients conferred increased risk of POAG onset. AT5, AT10, and AT14 were significantly associated with rapid VF progression. In a subgroup analysis by high-risk ATs (>95th percentile or <75th percentile coefficients), PSD lost significance as a predictor of POAG in the low-risk group., Conclusions: Baseline VFs, prior to detectable glaucomatous damage, contain occult patterns representing early changes that may increase the risk of POAG onset and VF progression in patients with ocular hypertension. The relationship between PSD and POAG is modified by the presence of high-risk patterns at baseline. An AT-based prediction model for POAG may provide more interpretable glaucoma-specific information in a clinical setting.

Purpose: To develop an automated deep learning system for detecting the presence and location of disc hemorrhages in optic disc photographs., Design: Development and testing of a deep learning algorithm., Methods: Optic disc photos (597 images with at least 1 disc hemorrhage and 1075 images without any disc hemorrhage from 1562 eyes) from 5 institutions were classified by expert graders based on the presence or absence of disc hemorrhage. The images were split into training (n = 1340), validation (n = 167), and test (n = 165) datasets. Two state-of-the-art deep learning algorithms based on either object-level detection or image-level classification were trained on the dataset. These models were compared to one another and against 2 independent glaucoma specialists. We evaluated model performance by the area under the receiver operating characteristic curve (AUC). AUCs were compared with the Hanley-McNeil method., Results: The object detection model achieved an AUC of 0.936 (95% CI = 0.857-0.964) across all held-out images (n = 165 photographs), which was significantly superior to the image classification model (AUC = 0.845, 95% CI = 0.740-0.912; P = .006). At an operating point selected for high specificity, the model achieved a specificity of 94.3% and a sensitivity of 70.0%, which was statistically indistinguishable from an expert clinician (P = .7). At an operating point selected for high sensitivity, the model achieves a sensitivity of 96.7% and a specificity of 73.3%., Conclusions: An autonomous object detection model is superior to an image classification model for detecting disc hemorrhages, and performed comparably to 2 clinicians., (Copyright © 2023 The Authors. Published by Elsevier Inc. All rights reserved.)

Ophthalmic images, along with their derivatives like retinal nerve fiber layer (RNFL) thickness maps, play a crucial role in detecting and monitoring eye diseases such as glaucoma. For computer-aided diagnosis of eye diseases, the key technique is to automatically extract meaningful features from ophthalmic images that can reveal the biomarkers (e.g., RNFL thinning patterns) associated with functional vision loss. However, representation learning from ophthalmic images that links structural retinal damage with human vision loss is non-trivial mostly due to large anatomical variations between patients. This challenge is further amplified by the presence of image artifacts, commonly resulting from image acquisition and automated segmentation issues. In this paper, we present an artifact-tolerant unsupervised learning framework called EyeLearn for learning ophthalmic image representations in glaucoma cases. EyeLearn includes an artifact correction module to learn representations that optimally predict artifact-free images. In addition, EyeLearn adopts a clustering-guided contrastive learning strategy to explicitly capture the affinities within and between images. During training, images are dynamically organized into clusters to form contrastive samples, which encourage learning similar or dissimilar representations for images in the same or different clusters, respectively. To evaluate EyeLearn, we use the learned representations for visual field prediction and glaucoma detection with a real-world dataset of glaucoma patient ophthalmic images. Extensive experiments and comparisons with state-of-the-art methods confirm the effectiveness of EyeLearn in learning optimal feature representations from ophthalmic images.

Background: Circumpapillary retinal nerve fiber layer thickness (cpRNFLT) as assessed by spectral domain optical coherence tomography (SD-OCT) is a new technique used for the detection and evaluation of glaucoma and other optic neuropathies. Before translating cpRNFLT into clinics, it is crucially important to investigate anthropometric, biochemical, and clinical parameters potentially affecting cpRNFLT in a large population-based dataset., Methods: The population-based LIFE-Adult Study randomly selected 10,000 participants from the population registry of Leipzig, Germany. All participants underwent standardized systemic assessment of various cardiometabolic risk markers and ocular imaging, including cpRNFLT measurement using SD-OCT (Spectralis, Heidelberg Engineering). After employing strict SD-OCT quality criteria, 8952 individuals were analyzed. Multivariable linear regression analyses were used to evaluate the independent associations of various cardiometabolic risk markers with sector-specific cpRNFLT. For significant markers, the relative strength of the observed associations was compared to each other to identify the most relevant factors influencing cpRNFLT. In all analyses, the false discovery rate method for multiple comparisons was applied., Results: In the entire cohort, female subjects had significantly thicker global and also sectoral cpRNFLT compared to male subjects (p < 0.05). Multivariable linear regression analyses revealed a significant and independent association between global and sectoral cpRNFLT with biomarkers of renal function and lipid profile. Thus, thinner cpRNFLT was associated with worse renal function as assessed by cystatin C and estimated glomerular filtration rate. Furthermore, an adverse lipid profile (i.e., low high-density lipoprotein (HDL) cholesterol, as well as high total, high non-HDL, high low-density lipoprotein cholesterol, and high apolipoprotein B) was independently and statistically significantly related to thicker cpRNFLT. In contrast, we do not observe a significant association between cpRNFLT and markers of inflammation, glucose homeostasis, liver function, blood pressure, or obesity in our sector-specific analysis and globally., Conclusions: Markers of renal function and lipid metabolism are predictors of sectoral cpRNFLT in a large and deeply phenotyped population-based study independently of previously established covariates. Future studies on cpRNFLT should include these biomarkers and need to investigate whether incorporation will improve the diagnosis of early eye diseases based on cpRNFLT., (© 2021. The Author(s).)

Purpose: To assess the structure-function relationship in glaucoma using Humphrey visual field (HVF) perimetry and a three-dimensional neuroretinal rim parameter derived from spectral domain optical coherence tomography (SD-OCT) volume scans., Methods: Structure-function correlation was analyzed globally and regionally (four quadrants and four sectors). Structural data included peripapillary retinal nerve fiber layer (RNFL) thickness and minimum distance band (MDB) neuroretinal rim thickness, defined as the shortest distance between the inner cup surface and the outer retinal pigment epithelium/Bruch's membrane complex. Logarithmic regression analyses were performed and Pearson correlation coefficients determined to assess relationship strength., Results: The study consisted of 102 open-angle glaucoma patients and 58 healthy subjects. The Pearson correlation coefficient for global MDB thickness (R = 0.585) was higher than for global RNFL thickness (R = 0.492), but the difference was not statistically significant (P = 0.18). The correlation coefficients for regional MDB thicknesses and corresponding HVF sensitivities were higher than those for regional RNFL thicknesses and HVF in six out of eight regions (P = 0.08 to 0.47). In the remaining two out of eight regions, the correlation coefficients were higher for RNFL thickness than for MDB thickness (P = 0.15 to 0.20)., Conclusions: Three-dimensional MDB neuroretinal rim thickness relates to visual function as strongly as the most commonly used SD-OCT parameter for glaucoma, two-dimensional peripapillary RNFL thickness., Translational Relevance: This paper illustrates the potential for 3D OCT algorithms to improve in vivo imaging in glaucoma.

Precis: In open-angle glaucoma, when neuroretinal rim tissue measured by volumetric optical coherence tomography (OCT) scans is below a third of the normal value, visual field (VF) damage becomes detectable., Purpose: To determine the amount of neuroretinal rim tissue thickness below which VF damage becomes detectable., Methods: In a retrospective cross-sectional study, 1 eye per subject (of 57 healthy and 100 open-angle glaucoma patients) at an academic institution had eye examinations, VF testing, spectral-domain OCT retinal nerve fiber layer (RNFL) thickness measurements, and optic nerve volumetric scans. Using custom algorithms, the minimum distance band (MDB) neuroretinal rim thickness was calculated from optic nerve scans. "Broken-stick" regression was performed for estimating both the MDB and RNFL thickness tipping-point thresholds, below which were associated with initial VF defects in the decibel scale. The slopes for the structure-function relationship above and below the thresholds were computed. Smoothing curves of the MDB and RNFL thickness covariates were evaluated to examine the consistency of the independently identified tipping-point pairs., Results: Plots of VF total deviation against MDB thickness revealed plateaus of VF total deviation unrelated to MDB thickness. Below the thresholds, VF total deviation decreased with MDB thickness, with the associated slopes significantly greater than those above the thresholds (P<0.014). Below 31% of global MDB thickness, and 36.8% and 43.6% of superior and inferior MDB thickness, VF damage becomes detectable. The MDB and RNFL tipping points were in good accordance with the correlation of the MDB and RNFL thickness covariates., Conclusions: When neuroretinal rim tissue, characterized by MDB thickness in OCT, is below a third of the normal value, VF damage in the decibel scale becomes detectable.

Purpose: The onset and progression of optic neuropathies like glaucoma often occurs asymmetrically between the two eyes of a patient. Interocular circumpapillary retinal nerve fiber layer thickness (cpRNFLT) differences could detect disease earlier. To apply such differences diagnostically, detailed location specific norms are necessary., Methods: Spectral-domain optical coherence tomography cpRNFLT circle scans from the population-based Leipzig Research Centre for Civilization Diseases-Adult study were selected. At each of the 768 radial scanning locations, normative interocular cpRNFLT difference distributions were calculated based on age and interocular radius difference., Results: A total of 8966 cpRNFLT scans of healthy eyes (4483 patients; 55% female; age range, 20-79 years) were selected. Global cpRNFLT average was 1.53 µm thicker in right eyes ( P < 2.2 × 10 -16 ). On 96% of the 768 locations, left minus right eye differences were significant ( P < 0.05), varying between +11.6 µm (superonasal location) and -11.8 µm (nasal location). Increased age and difference in interocular scanning radii were associated with an increased mean and variance of interocular cpRNFLT difference at most retinal locations, apart from the area temporal to the inferior RNF bundle where cpRNFLT becomes more similar between eyes with age., Conclusions: We provide pointwise normative distributions of interocular cpRNFLT differences at an unprecedentedly high spatial resolution of 768 A-scans and reveal considerable location specific asymmetries as well as their associations with age and scanning radius differences between eyes., Translational Relevance: To facilitate clinical application, we implement these age- and radius-specific norms across all 768 locations in an open-source software to generate patient-specific normative color plots., Competing Interests: Disclosure: N. Baniasadi, None; F.G. Rauscher, None; D. Li, None; M. Wang, None; E.Y. Choi, None; H. Wang, None; T. Peschel, None; K. Wirkner, None; T. Kirsten, None; J. Thiery, None; C. Engel, None; M. Loeffler, None; T. Elze, None, (Copyright 2020 The Authors.)

Purpose: To investigate systemic and ocular determinants of peripapillary retinal nerve fiber layer thickness (pRNFLT) in the European population., Design: Cross-sectional meta-analysis., Participants: A total of 16 084 European adults from 8 cohort studies (mean age range, 56.9±12.3-82.1±4.2 years) of the European Eye Epidemiology (E3) consortium., Methods: We examined associations with pRNFLT measured by spectral-domain OCT in each study using multivariable linear regression and pooled results using random effects meta-analysis., Main Outcome Measures: Determinants of pRNFLT., Results: Mean pRNFLT ranged from 86.8±21.4 μm in the Rotterdam Study I to 104.7±12.5 μm in the Rotterdam Study III. We found the following factors to be associated with reduced pRNFLT: Older age (β = -0.38 μm/year; 95% confidence interval [CI], -0.57 to -0.18), higher intraocular pressure (IOP) (β = -0.36 μm/mmHg; 95% CI, -0.56 to -0.15), visual impairment (β = -5.50 μm; 95% CI, -9.37 to -1.64), and history of systemic hypertension (β = -0.54 μm; 95% CI, -1.01 to -0.07) and stroke (β = -1.94 μm; 95% CI, -3.17 to -0.72). A suggestive, albeit nonsignificant, association was observed for dementia (β = -3.11 μm; 95% CI, -6.22 to 0.01). Higher pRNFLT was associated with more hyperopic spherical equivalent (β = 1.39 μm/diopter; 95% CI, 1.19-1.59) and smoking (β = 1.53 μm; 95% CI, 1.00-2.06 for current smokers compared with never-smokers)., Conclusions: In addition to previously described determinants such as age and refraction, we found that systemic vascular and neurovascular diseases were associated with reduced pRNFLT. These may be of clinical relevance, especially in glaucoma monitoring of patients with newly occurring vascular comorbidities., (Copyright © 2018 American Academy of Ophthalmology. Published by Elsevier Inc. All rights reserved.)

We studied the relationship between major retinal blood vessel (BV) positions and glaucoma parameters based on pairs of Cirrus optical coherence tomography scans and Humphrey visual fields of 445 eyes from 445 glaucoma patients in our cross-sectional study. A trained observer marked the major superior and inferior temporal BV (artery and vein) positions on four concentric circles around the optic disc. Analysis of variance was performed to analyze the group differences of BV positions related to the factors of radius, BV type, myopia status and glaucoma stage. Subsequent t-tests were implemented to further study the effect of glaucoma stage on BV positions. The radial variations of BV positions were correlated to mean deviation and circumpapillary retinal nerve fiber layer thickness (cpRNFLT). We found significant main effects of BV type, radius and myopia status for superior and inferior BV positions and of glaucoma stage for superior BV positions (all p≤0.006) with significant superior artery nasalization in advanced compared to mild glaucoma on the two smallest circles (subsequent t-tests, p<0.05). In addition, MD (r = -0.10, p = 0.04) and cpRNFLT (r = -0.12, p = 0.02) were significantly correlated to the angle difference of superior arteries between the innermost and outermost circles. In conclusion, we demonstrated that peripapillary superior artery positions are significantly nasalized for advanced glaucoma.

Clinicians use retinal nerve fiber layer thickness (RNFLT) measured by optical coherence tomography (OCT) as an adjunct to glaucoma diagnosis. Ametropia is accompanied by changes to the optic nerve head (ONH), which may affect how OCT machines mark RNFLT measurements as abnormal. These changes in abnormality patterns may bias glaucoma diagnosis. Here, we investigate the relationship between OCT abnormality patterns and the following ONH-related and ametropia-associated parameters on 421 eyes of glaucoma patients: optic disc tilt and torsion, central retinal vessel trunk location (CRVTL), and nasal and temporal retinal curvature adjacent to ONH, quantified as nasal/temporal slopes of the inner limiting membrane. We applied multivariate logistic regression with abnormality marks as regressands to 40,401 locations of the peripapillary region and generated spatial maps of locations of false positive/negative abnormality marks independent of glaucoma severity. Effects of torsion and temporal slope were negligible. The effect of tilt could be explained by covariation with ametropia. For CRVTL/nasal slope, abnormality pattern shifts at 7.2%/23.5% of the peripapillary region were detected, respectively, independent of glaucoma severity and ametropia. Therefore, CRVTL and nasal curvature should be included in OCT RNFLT norms. Our spatial location maps may aid clinicians to improve diagnostic accuracy., ((2017) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE).)

Purpose: To assess the clinical validity of visual field (VF) archetypal analysis, a previously developed machine learning method for decomposing any Humphrey VF (24-2) into a weighted sum of clinically recognizable VF loss patterns., Materials and Methods: For each of 16 previously identified VF loss patterns ("archetypes," denoted AT1 through AT16), we screened 30,995 reliable VFs to select 10-20 representative patients whose VFs had the highest decomposition coefficients for each archetype. VF global indices and patient ocular and demographic features were extracted retrospectively. Based on resemblances between VF archetypes and clinically observed VF patterns, hypotheses were generated for associations between certain VF archetypes and clinical features, such as an association between AT6 (central island, representing severe VF loss) and large cup-to-disk ratio (CDR). Distributions of the selected clinical features were compared between representative eyes of certain archetypes and all other eyes using the two-tailed t-test or Fisher exact test., Results: 243 eyes from 243 patients were included, representative of AT1 through AT16. CDR was more often ≥ 0.7 among eyes representative of AT6 (central island; p = 0.002), AT10 (inferior arcuate defect; p = 0.048), AT14 (superior paracentral defect; p = 0.016), and AT16 (inferior paracentral defect; p = 0.016) than other eyes. CDR was more often < 0.7 among eyes representative of AT1 (no focal defect; p < 0.001) and AT2 (superior defect; p = 0.027), which was also associated with ptosis (p < 0.001). AT12 (temporal hemianopia) was associated with history of stroke (p = 0.022). AT11 (concentric peripheral defect) trended toward association with trial lens correction > 6D (p = 0.069)., Conclusions: Shared clinical features between computationally derived VF archetypes and clinically observed VF patterns support the clinical validity of VF archetypal analysis.