Artificial Intelligence Classification of Central Visual Field Patterns in Glaucoma

Contributions of Brain Microstructures and Metabolism to Visual Field Loss Patterns in Glaucoma Using Archetypal and Information Gain Analyses

Purpose

To investigate the contributions of the microstructural and metabolic brain environment to glaucoma and their association with visual field (VF) loss patterns by using advanced diffusion magnetic resonance imaging (dMRI), proton magnetic resonance spectroscopy (MRS), and clinical ophthalmic measures.

Methods

Sixty-nine glaucoma and healthy subjects underwent dMRI and/or MRS at 3 Tesla. Ophthalmic data were collected from VF perimetry and optical coherence tomography. dMRI parameters of microstructural integrity in the optic radiation and MRS-derived neurochemical levels in the visual cortex were compared among early glaucoma, advanced glaucoma, and healthy controls. Multivariate regression was used to correlate neuroimaging metrics with 16 archetypal VF loss patterns. We also ranked neuroimaging, ophthalmic, and demographic attributes in terms of their information gain to determine their importance to glaucoma.

Results

In dMRI, decreasing fractional anisotropy, radial kurtosis, and tortuosity and increasing radial diffusivity correlated with greater overall VF loss bilaterally. Regionally, decreasing intra-axonal space and extra-axonal space diffusivities correlated with greater VF loss in the superior-altitudinal area of the right eye and the inferior-altitudinal area of the left eye. In MRS, both early and advanced glaucoma patients had lower gamma-aminobutyric acid (GABA), glutamate, and choline levels than healthy controls. GABA appeared to associate more with superonasal VF loss, and glutamate and choline more with inferior VF loss. Choline ranked third for importance to early glaucoma, whereas radial kurtosis and GABA ranked fourth and fifth for advanced glaucoma.

Conclusions

Our findings highlight the importance of non-invasive neuroimaging biomarkers and analytical modeling for unveiling glaucomatous neurodegeneration and how they reflect complementary VF loss patterns.

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

Purpose

To develop an objective glaucoma damage severity classification system based on OCT-derived retinal nerve fiber layer (RNFL) thickness measurements.

Design

Algorithm development for RNFL damage severity classification based on multicenter OCT data.

Subjects and Participants

A total of 6561 circumpapillary RNFL profiles from 2269 eyes of 1171 subjects to develop models, and 2505 RNFL profiles from 1099 eyes of 900 subjects to validate models.

Methods

We developed an unsupervised k-means model to identify clusters of eyes with similar RNFL thickness profiles. We annotated the clusters based on their respective global RNFL thickness. We computed the optimal global RNFL thickness thresholds that discriminated different severity levels based on Bayes' minimum error principle. We validated the proposed pipeline based on an independent validation dataset with 2505 RNFL profiles from 1099 eyes of 900 subjects.

Main Outcome Measures

Accuracy, area under the receiver operating characteristic curve, and confusion matrix.

Results

The k-means clustering discovered 4 clusters with 1382, 1613, 1727, and 1839 samples with mean (standard deviation) global RNFL thickness of 58.3 (8.9) μm, 78.9 (6.7) μm, 87.7 (8.2) μm, and 101.5 (7.9) μm. The Bayes' minimum error classifier identified optimal global RNFL values of > 95 μ m , 86 to 95 μ m , 70 to 85 μ m , and < 70 μ m for discriminating normal eyes and eyes at the early, moderate, and advanced stages of RNFL thickness loss, respectively. About 4% of normal eyes and 98% of eyes with advanced RNFL loss had either global, or ≥ 1 quadrant, RNFL thickness outside of normal limits provided by the OCT instrument.

Conclusions

Unsupervised machine learning discovered that the optimal RNFL thresholds for separating normal eyes and eyes with early, moderate, and advanced RNFL loss were 95 μ m , 85 μm, and 70 μ m , respectively. This RNFL loss classification system is unbiased as there was no preassumption or human expert intervention in the development process. Additionally, it is objective, easy to use, and consistent, which may augment glaucoma research and day-to-day clinical practice.

Financial Disclosures

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Machine Learning-Derived Baseline Visual Field Patterns Predict Future Glaucoma Onset in the Ocular Hypertension Treatment Study

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.

Artificial intelligence in glaucoma: opportunities, challenges, and future directions

Artificial intelligence (AI) has shown excellent diagnostic performance in detecting various complex problems related to many areas of healthcare including ophthalmology. AI diagnostic systems developed from fundus images have become state-of-the-art tools in diagnosing retinal conditions and glaucoma as well as other ocular diseases. However, designing and implementing AI models using large imaging data is challenging. In this study, we review different machine learning (ML) and deep learning (DL) techniques applied to multiple modalities of retinal data, such as fundus images and visual fields for glaucoma detection, progression assessment, staging and so on. We summarize findings and provide several taxonomies to help the reader understand the evolution of conventional and emerging AI models in glaucoma. We discuss opportunities and challenges facing AI application in glaucoma and highlight some key themes from the existing literature that may help to explore future studies. Our goal in this systematic review is to help readers and researchers to understand critical aspects of AI related to glaucoma as well as determine the necessary steps and requirements for the successful development of AI models in glaucoma.

Uncertainty-inspired open set learning for retinal anomaly identification

Failure to recognize samples from the classes unseen during training is a major limitation of artificial intelligence in the real-world implementation for recognition and classification of retinal anomalies. We establish an uncertainty-inspired open set (UIOS) model, which is trained with fundus images of 9 retinal conditions. Besides assessing the probability of each category, UIOS also calculates an uncertainty score to express its confidence. Our UIOS model with thresholding strategy achieves an F1 score of 99.55%, 97.01% and 91.91% for the internal testing set, external target categories (TC)-JSIEC dataset and TC-unseen testing set, respectively, compared to the F1 score of 92.20%, 80.69% and 64.74% by the standard AI model. Furthermore, UIOS correctly predicts high uncertainty scores, which would prompt the need for a manual check in the datasets of non-target categories retinal diseases, low-quality fundus images, and non-fundus images. UIOS provides a robust method for real-world screening of retinal anomalies.

Deep learning model to identify homonymous defects on automated perimetry

BACKGROUND

Homonymous visual field (VF) defects are usually an indicator of serious intracranial pathology but may be subtle and difficult to detect. Artificial intelligence (AI) models could play a key role in simplifying the detection of these defects. This study aimed to develop an automated deep learning AI model to accurately identify homonymous VF defects from automated perimetry.

METHODS

VFs performed on Humphrey field analyser (24-2 algorithm) were collected and run through an in-house optical character recognition program that extracted mean deviation data and prepared it for use in the proposed AI model. The deep learning AI model, Deep Homonymous Classifier, was developed using PyTorch framework and used convolutional neural networks to extract spatial features for binary classification. Total collected dataset underwent 7-fold cross validation for model training and evaluation. To address dataset class imbalance, data augmentation techniques and state-of-the-art loss function that uses complement cross entropy were used to train and enhance the proposed AI model.

RESULTS

The proposed model was evaluated using 7-fold cross validation and achieved an average accuracy of 87% for detecting homonymous VF defects in previously unseen VFs. Recall, which is a critical value for this model as reducing false negatives is a priority in disease detection, was found to be on average 92%. The calculated F2 score for the proposed model was 0.89 with a Cohen's kappa value of 0.70.

CONCLUSION

This newly developed deep learning model achieved an overall average accuracy of 87%, making it highly effective in identifying homonymous VF defects on automated perimetry.

Archetypal analysis of longitudinal visual fields for idiopathic intracranial hypertension patients presenting in a clinic setting

We previously applied archetypal analysis (AA) using visual fields (VF) from the Idiopathic Intracranial Hypertension Treatment Trial (IIHTT) to derive a model, which quantified patterns (or archetypes [ATs] of VF loss), anticipated recovery, and identified residual VF deficits. We hypothesized that AA could produce similar results using IIH VFs collected in clinical practice. We applied AA to 803 VFs from 235 eyes with IIH from an outpatient neuro-ophthalmology clinic and created a clinic-derived model of ATs, with the relative weight (RW) and average total deviation (TD) for each AT. We also created a combined-derived model from an input dataset containing the clinic VFs and 2862 VFs from the IIHTT. We used both models to decompose clinic VF into ATs of varying percent weight (PW), correlated presentation AT PW with mean deviation (MD), and evaluated final visit VFs considered "normal" by MD ≥ -2.00 dB for residual abnormal ATs. The 14-AT clinic-derived and combined-derived models revealed similar patterns of VF loss previously identified in the IIHTT model. AT1 (a normal pattern) was most prevalent in both models (RW = 51.8% for clinic-derived; 35.4% for combined-derived). Presentation AT1 PW correlated with final visit MD (r = 0.82, p < 0.001 for the clinic-derived model; r = 0.59, p < 0.001 for the combined-derived model). Both models showed ATs with similar patterns of regional VF loss. The most common patterns of VF loss in "normal" final visit VFs using each model were clinic-derived AT2 (mild global depression with enlarged blind spot; 44/125 VFs; 34%) and combined-derived AT2 (near-normal; 93/149 VFs; 62%). AA provides quantitative values for IIH-related patterns of VF loss that can be used to monitor VF changes in a clinic setting. Presentation AT1 PW is associated with the degree of VF recovery. AA identifies residual VF deficits not otherwise indicated by MD.

Applications of Artificial Intelligence and Deep Learning in Glaucoma

Diagnosis and detection of progression of glaucoma remains challenging. Artificial intelligence-based tools have the potential to improve and standardize the assessment of glaucoma but development of these algorithms is difficult given the multimodal and variable nature of the diagnosis. Currently, most algorithms are focused on a single imaging modality, specifically screening and diagnosis based on fundus photos or optical coherence tomography images. Use of anterior segment optical coherence tomography and goniophotographs is limited. The majority of algorithms designed for disease progression prediction are based on visual fields. No studies in our literature search assessed the use of artificial intelligence for treatment response prediction and no studies conducted prospective testing of their algorithms. Additional challenges to the development of artificial intelligence-based tools include scarcity of data and a lack of consensus in diagnostic criteria. Although research in the use of artificial intelligence for glaucoma is promising, additional work is needed to develop clinically usable tools.

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

PURPOSE

To test the hypothesis that macular ganglion cell layer (GCL) measurements detect early glaucoma with higher accuracy than ganglion cell/inner plexiform layer (GCIPL) thickness measurements.

DESIGN

Cross-sectional study.

PARTICIPANTS

The first cohort included 58 glaucomatous eyes with visual field mean deviation (MD) ≥ -6 dB and 125 normal eyes. The second cohort included 72 glaucomatous and 73 normal/glaucoma suspect (GS) eyes with scans able to create GCL/GCIPL deviation maps.

METHODS

In the first cohort, 8 × 8 GCL and GCIPL grids were exported and 5 superior and inferior sectors were defined. Global and sectoral GCL and GCIPL measures were used to predict glaucoma. In the second cohort, proportions of scan areas with abnormal (< 5% and < 1% cutoffs) and supernormal (> 95% and > 99% cutoffs) thicknesses on deviation maps were calculated. The extents of GCL and GCIPL abnormal areas were used to predict glaucoma.

MAIN OUTCOME MEASURES

Extents of abnormal GCL/GCIPL regions and areas under receiver operating characteristic curves (AUROC) for prediction of glaucoma were compared between GCL or GCIPL measures.

RESULTS

The average ± standard deviation MDs were -3.7 ± 1.6 dB and -2.7 ± 1.8 dB in glaucomatous eyes in the first and second cohorts, respectively. Global GCIPL thickness measures (central 18° × 18° macular region) performed better than GCL for early detection of glaucoma (AUROC, 0.928 vs. 0.884, respectively; P = 0.004). Superior and inferior sector 3 thickness measures provided the best discrimination with both GCL and GCIPL (inferior GCL AUROC, 0.860 vs. GCIPL AUROC, 0.916 [P = 0.001]; superior GCL AUROC, 0.916 vs. GCIPL AUROC, 0.900 [P = 0.24]). The extents of abnormal GCL regions at a 1% cutoff in the central elliptical area were 17.5 ± 22.2% and 6.4 ± 10.8% in glaucomatous and normal/GS eyes, respectively, versus 17.0 ± 22.2% and 5.7 ± 10.5%, respectively, for GCIPL (P = 0.06 for GCL and 0.002 for GCIPL). The extents of GCL and GCIPL supernormal regions were mostly similar in glaucomatous and normal eyes. The best performance for prediction of glaucoma in the second cohort was detected at a P value of < 1% within the entire scan for both GCL and GCIPL (AUC, 0.681 vs. 0.668, respectively; P = 0.29).

CONCLUSIONS

Macular GCL and GCIPL thicknesses are equivalent for identifying early glaucoma with current OCT technology. This is likely explained by limitations of inner macular layer segmentation and concurrent changes within the inner plexiform layer in early glaucoma.

Machine-Identified Patterns of Visual Field Loss and an Association with Rapid Progression in the Ocular Hypertension Treatment Study

PURPOSE

To identify patterns of visual field (VF) loss based on unsupervised machine learning and to identify patterns that are associated with rapid progression.

DESIGN

Cross-sectional and longitudinal study.

PARTICIPANTS

A total of 2231 abnormal VFs from 205 eyes of 176 Ocular Hypertension Treatment Study (OHTS) participants followed over approximately 16 years.

METHODS

Visual fields were assessed by an unsupervised deep archetypal analysis algorithm and an OHTS-certified VF reader to identify prevalent patterns of VF loss. Machine-identified patterns of glaucoma damage were compared against those patterns previously identified (expert-identified) in the OHTS in 2003. Based on the longitudinal VFs of each eye, VF loss patterns that were strongly associated with rapid glaucoma progression were identified.

MAIN OUTCOME MEASURES

Machine-expert correspondence and type of patterns of VF loss associated with rapid progression.

RESULTS

The average VF mean deviation (MD) at conversion to glaucoma was -2.7 decibels (dB) (standard deviation [SD] = 2.4 dB), whereas the average MD of the eyes at the last visit was -5.2 dB (SD = 5.5 dB). Fifty out of 205 eyes had MD rate of -1 dB/year or worse and were considered rapid progressors. Eighteen machine-identified patterns of VF loss were compared with expert-identified patterns, in which 13 patterns of VF loss were similar. The most prevalent expert-identified patterns included partial arcuate, paracentral, and nasal step defects, and the most prevalent machine-identified patterns included temporal wedge, partial arcuate, nasal step, and paracentral VF defects. One of the machine-identified patterns of VF loss predicted future rapid VF progression after adjustment for age, sex, and initial MD.

CONCLUSIONS

An automated machine learning system can identify patterns of VF loss and could provide objective and reproducible nomenclature for characterizing early signs of visual defects and rapid progression in patients with glaucoma.

The number of examinations required for the accurate prediction of the progression of the central 10-degree visual field test in glaucoma

The purpose of the study was to investigate the number of examinations required to precisely predict the future central 10-degree visual field (VF) test and to evaluate the effect of fitting non-linear models, including quadratic regression, exponential regression, logistic regression, and M-estimator robust regression model, for eyes with glaucoma. 180 eyes from 133 open angle glaucoma patients with a minimum of 13 Humphrey Field Analyzer 10-2 SITA standard VF tests were analyzed in this study. Using trend analysis with ordinary least squares linear regression (OLSLR), the first, second, and third future VFs were predicted in a point-wise (PW) manner using a varied number of prior VF sequences, and mean absolute errors (MAE) were calculated. The number of VFs needed to reach the minimum 95% confidence interval (CI) of the MAE of the OLSLR was investigated. We also examined the effect of applying other non-linear models. When predicting the first, second, and third future VFs using OLSLR, the minimum MAE was obtained using VF1-12 (2.15 ± 0.98 dB), VF1-11 (2.33 ± 1.10 dB), and VF1-10 (2.63 ± 1.36 dB), respectively. To reach the 95% CI of these MAEs, 10, 10, and 8 VFs were needed for the first, second and third future VF predictions, respectively. No improvement was observed by applying non-linear regression models. As a conclusion, approximately 8-10 VFs were needed to achieve an accurate prediction of PW VF sensitivity of the 10-degree central VF.

Archetypal analysis of visual fields in optic neuritis reveals functional biomarkers associated with outcome and treatment response

BACKGROUND AND OBJECTIVES

Archetypal analysis (AA), a form of unsupervised machine learning, can identify quantifiable visual field (VF) patterns seen in optic neuritis (ON), known as archetypes (ATs). We hypothesized that AT weight changes over time would reflect the course of recovery and the effects of therapy in ON. We explored whether baseline AT weights would be associated with VF status at the clinical trial outcome and if ATs would indicate residual VF defects in eyes with mean deviation (MD) ≥ -2.00 at six months.

METHODS

We used a published 16-AT model derived from 3892 Optic Neuritis Treatment Trial VFs (456 eyes) for all analyses. We measured AT weight changes over the six-month study period and used asymptotic regression to analyze the rate of change. We compared AT weights at six months between treatment groups. We evaluated associations between baseline AT weight thresholds and VF outcome or treatment effect. We calculated residual AT weights in eyes with MD ≥ -2.00 dB at six months.

RESULTS

Over six months, AT1 (a normal VF pattern) demonstrated the greatest median weight change, increasing from 0.00% (IQR 0.00-0.00%) at baseline to 60.0% (IQR 38.3-70.8%) at six months (p < 0.001). At outcome, the intravenous methylprednisolone (IVMP) group had the highest median AT1 weight (IVMP: 63.3%, IQR 51.3-72.8%; placebo: 56.2%, IQR 35.1-71.6%; prednisone 58.3%, IQR 35.1-71.6%; p = 0.019). Eyes with AT1 weight ≥ 19% at baseline had superior median MD values (-0.91 vs. -2.07 dB, p < 0.001) and AT1 weights (70.8% vs. 57.8% p < 0.001) at six months. Only eyes with AT1 weight < 19% at baseline showed a treatment benefit for IVMP, with a higher six-month median AT1 weight compared to placebo (p = 0.015) and prednisone (p = 0.016), and a higher median MD compared to placebo (p = 0.027). At six months, 182 (80.2%) VFs with MD ≥ -2.00 had at least one abnormal AT.

DISCUSSION

Changes in quantifiable, archetypal patterns of VF loss reflect recovery in ON. Machine learning analysis of the VFs in optic neuritis reveals associations with response to therapy and VF outcome, and uncovers residual deficits, not readily seen with standard evaluations.

Longitudinal visual field archetypal analysis of optic neuritis treated in a clinical setting

Background/aims

We previously used archetypal analysis (AA) to create a model that quantified patterns (archetypes (ATs)) of visual field (VF) loss that can predict recovery and reveal residual VF deficits from eyes in the Optic Neuritis Treatment Trial (ONTT). We hypothesised that AA could produce similar results for ON VFs collected in clinical practice.

Methods

We applied AA to a retrospective data set of 486 VFs collected in 1 neuro-ophthalmology service from 141 eyes with acute ON and typical VF defects, to create a clinic-derived AT model. We also used the ONTT-derived AT model to analyse this new dataset. We compared the findings of both models by decomposing VFs into component ATs of varying per cent weight (PW), correlating presentation AT PW with mean deviation (MD) at final visits for each eye and identifying residual deficits in VFs considered normal.

Results

Both models, each with 16 ATs, decomposed each presentation VF into 0–6 abnormal ATs representative of known patterns of ON-related VF loss. AT1, the normal pattern in both models, correlated strongly with MD for VFs collected at presentation (r=0.82; p<0.001) and the final visit (r=0.81, p<0.001). The presentation AT1 PW was associated with improvement in MD over time. 67% of VFs considered ‘normal’ at final visit had 1.2±0.4 abnormal ATs, and both models revealed similar patterns of regional VF loss.

Conclusions

AA is a quantitative method to measure change and outcome of ON VFs. Presentation AT features are associated with MD at final visit. AA identifies residual VF deficits not otherwise indicated by MD.

A Prospective Longitudinal Study to Investigate Corneal Hysteresis as a Risk Factor of Central Visual Field Progression in Glaucoma

PURPOSE

To evaluate the role of corneal hysteresis (CH) as a risk factor of central visual field (VF) progression in a cohort of glaucoma suspect and glaucoma patients.

DESIGN

Prospective cohort study.

METHODS

Two hundred forty-eight eyes of 143 subjects who were followed for an average of 4.8 years with a minimum of 5 visits with 10-2 and 24-2 VF tests were included. Univariable and multivariable linear mixed-effects models were used to identify characteristics associated with the rate of change over time in 10-2 and 24-2 mean deviation (MD). Mixed-effects logistic regression was used to evaluate characteristics associated with an increased likelihood of event-based 10-2 VF progression based on the clustered pointwise linear regression criterion.

RESULTS

CH was significantly associated with 10-2 and 24-2 VF progression in the univariable trend-based analysis. In multivariable trend-based analyses, lower CH was associated with a faster rate of decline in 10-2 MD (0.07 dB/y per 1 mm Hg, P < .001) but not with 24-2 MD (P = .490). In multivariable event-based analysis, lower CH was associated with an increased likelihood of 10-2 VF progression (odds ratio = 1.35 per 1 mm Hg lower, P = .025). Similar results were found in eyes with early glaucomatous damage at the baseline (baseline: 24-2 MD ≥ -6 dB).

CONCLUSIONS

Lower CH was associated with a statistically significant, but relatively small, increased risk of central VF progression on the 10-2 test grid. Given the substantial influence of central VF impairment on the quality of life, clinicians should consider using CH to assess the risk of progression in patients with primary open-angle glaucoma including those with early disease.

An Objective and Easy-to-Use Glaucoma Functional Severity Staging System Based on Artificial Intelligence

OBJECTIVE

The objective of this study was to develop an objective and easy-to-use glaucoma staging system based on visual fields (VFs).

SUBJECTS AND PARTICIPANTS

A total of 13,231 VFs from 8077 subjects were used to develop models and 8024 VFs from 4445 subjects were used to validate models.

METHODS

We developed an unsupervised machine learning model to identify clusters with similar VF values. We annotated the clusters based on their respective mean deviation (MD). We computed optimal MD thresholds that discriminate clusters with the highest accuracy based on Bayes minimum error principle. We evaluated the accuracy of the staging system and validated findings based on an independent validation dataset.

RESULTS

The unsupervised k -means algorithm discovered 4 clusters with 6784, 4034, 1541, and 872 VFs and average MDs of 0.0 dB (±1.4: SD), -4.8 dB (±1.9), -12.2 dB (±2.9), and -23.0 dB (±3.8), respectively. The supervised Bayes minimum error classifier identified optimal MD thresholds of -2.2, -8.0, and -17.3 dB for discriminating normal eyes and eyes at the early, moderate, and advanced stages of glaucoma. The accuracy of the glaucoma staging system was 94%, based on identified MD thresholds with respect to the initial k -means clusters.

CONCLUSIONS

We discovered that 4 severity levels based on MD thresholds of -2.2, -8.0, and -17.3 dB, provides the optimal number of severity stages based on unsupervised and supervised machine learning. This glaucoma staging system is unbiased, objective, easy-to-use, and consistent, which makes it highly suitable for use in glaucoma research and for day-to-day clinical practice.

Patterns of Visual Field Loss in Early, Moderate, and Severe Stages of Open Angle Glaucoma

PRCIS

Even in the early stages of glaucomatous visual field defects (VFDs), 49% of the defects occurred in both hemifields and 28% involved the central 5 degrees of the visual field (VF), which may have prognostic values.

PURPOSE

The aim was to determine the patterns of glaucomatous VFDs in early, moderate and severe stages of primary open angle glaucoma, using the Glaucoma Staging Application.

METHODS

According to the Modified University of Sao Paulo Glaucoma Visual Field Staging System Classification, patients with early, moderate and advanced VFDs were selected by the Glaucoma Staging Application using all databases of the Humphrey Visual Field Analyser of a glaucoma referral practice. We analyzed one VF of the 100 patients included in each group. The analysis consisted of classifying all exams regarding the location of the defects, the hemifields involved, and the connection to the blind spot.

RESULTS

We analyzed 300 VF. In the Early group, 27% of the VFDs are connected to the physiological blind spot, 64% in the Moderate group, and 95% in the Severe group ( P <0.01). In the Early group, 28% of the defects involved the central 5 degrees of the fixation, 59% in the Moderate and 88% in the Severe group. In the Early group, 49% of the defects involved both hemifields, 80% in the Moderate and 80% in the Severe group.

CONCLUSION

With increasing glaucoma severity, VFD showed a more central pattern, connected to the blind spot, and involved both hemifields. In early disease, both hemifields were commonly affected and more than a quarter of VFD involved the central 5 degrees close to fixation. Careful monitoring of the central VF in glaucoma is suggested.

Archetypal Analysis Reveals Quantifiable Patterns of Visual Field Loss in Optic Neuritis

Purpose

Identifying and monitoring visual field (VF) defects due to optic neuritis (ON) relies on qualitative clinician interpretation. Archetypal analysis (AA), a form of unsupervised machine learning, is used to quantify VF defects in glaucoma. We hypothesized that AA can identify quantifiable, ON-specific patterns (as archetypes [ATs]) of VF loss that resemble known ON VF defects.

Methods

We applied AA to a dataset of 3892 VFs prospectively collected from 456 eyes in the Optic Neuritis Treatment Trial (ONTT), and decomposed each VF into component ATs (total weight = 100%). AA of 568 VFs from 61 control eyes was used to define a minimum meaningful (≤7%) AT weight and weight change. We correlated baseline ON AT weights with global VF indices, visual acuity, and contrast sensitivity. For eyes with a dominant AT (weight ≥50%), we compared the ONTT VF classification with the AT pattern.

Results

AA generated a set of 16 ATs containing patterns seen in the ONTT. These were distinct from control ATs. Baseline study eye VFs were decomposed into 2.9 ± 1.5 ATs. AT2, a global dysfunction pattern, had the highest mean weight at baseline (36%; 95% confidence interval, 33%–40%), and showed the strongest correlation with MD (r = −0.91; P < 0.001), visual acuity (r = 0.70; P < 0.001), and contrast sensitivity (r = −0.77; P < 0.001). Of 191 baseline VFs with a dominant AT, 81% matched the descriptive classifications.

Conclusions

AA identifies and quantifies archetypal, ON-specific patterns of VF loss.

Translational Relevance

AA is a quantitative, objective method for demonstrating and monitoring change in regional VF deficits in ON.

Artificial Intelligence, Heuristic Biases, and the Optimization of Health Outcomes: Cautionary Optimism

The use of artificial intelligence (AI) and machine learning (ML) in clinical care offers great promise to improve patient health outcomes and reduce health inequity across patient populations. However, inherent biases in these applications, and the subsequent potential risk of harm can limit current use. Multi-modal workflows designed to minimize these limitations in the development, implementation, and evaluation of ML systems in real-world settings are needed to improve efficacy while reducing bias and the risk of potential harms. Comprehensive consideration of rapidly evolving AI technologies and the inherent risks of bias, the expanding volume and nature of data sources, and the evolving regulatory landscapes, can contribute meaningfully to the development of AI-enhanced clinical decision making and the reduction in health inequity.

Variability and Power to Detect Progression of Different Visual Field Patterns

PURPOSE

To compare the variability and ability to detect visual field (VF) progression of 24-2, central 12 locations of the 24-2 and 10-2 VF tests in eyes with abnormal VFs.

DESIGN

Retrospective, multisite cohort.

PARTICIPANTS

A total of 52 806 24-2 and 11 966 10-2 VF tests from 7307 eyes from the Glaucoma Research Network database were analyzed. Only eyes with ≥ 5 visits and ≥ 2 years of follow-up were included.

METHODS

Linear regression models were used to calculate the rates of mean deviation (MD) change (slopes), whereas their residuals were used to assess variability across the entire MD range. Computer simulations (n = 10 000) based on real MD residuals of our sample were performed to estimate power to detect significant progression (P < 5%) at various rates of MD change.

MAIN OUTCOME MEASURES

Time required to detect progression.

RESULTS

For all 3 patterns, the MD variability was highest within the -5 to -20 decibel (dB) range and consistently lower with the 10-2 compared with 24-2 or central 24-2. Overall, time to detect confirmed significant progression at 80% power was the lowest with 10-2 VF, with a decrease of 14.6% to 18.5% when compared with 24-2 and a decrease of 22.9% to 26.5% when compared with central 24-2.

CONCLUSIONS

Time to detect central VF progression was reduced with 10-2 MD compared with 24-2 and C24-2 MD in glaucoma eyes in this large dataset, in part because 10-2 tests had lower variability. These findings contribute to current evidence of the potential value of 10-2 testing in the clinical management of patients with glaucoma and in clinical trial design.

Updates in deep learning research in ophthalmology

Ophthalmology has been one of the early adopters of artificial intelligence (AI) within the medical field. Deep learning (DL), in particular, has garnered significant attention due to the availability of large amounts of data and digitized ocular images. Currently, AI in Ophthalmology is mainly focused on improving disease classification and supporting decision-making when treating ophthalmic diseases such as diabetic retinopathy, age-related macular degeneration (AMD), glaucoma and retinopathy of prematurity (ROP). However, most of the DL systems (DLSs) developed thus far remain in the research stage and only a handful are able to achieve clinical translation. This phenomenon is due to a combination of factors including concerns over security and privacy, poor generalizability, trust and explainability issues, unfavorable end-user perceptions and uncertain economic value. Overcoming this challenge would require a combination approach. Firstly, emerging techniques such as federated learning (FL), generative adversarial networks (GANs), autonomous AI and blockchain will be playing an increasingly critical role to enhance privacy, collaboration and DLS performance. Next, compliance to reporting and regulatory guidelines, such as CONSORT-AI and STARD-AI, will be required to in order to improve transparency, minimize abuse and ensure reproducibility. Thirdly, frameworks will be required to obtain patient consent, perform ethical assessment and evaluate end-user perception. Lastly, proper health economic assessment (HEA) must be performed to provide financial visibility during the early phases of DLS development. This is necessary to manage resources prudently and guide the development of DLS.

Demographic, Comorbid, and Clinical Variables Associated With Pointwise Visual Field Damage in Glaucoma: Data From the AGIS and CIGTS Clinical Trials

Purpose

To investigate differences across the visual field (VF) in the rate of glaucomatous progression, the likelihood of defect in four disease severity cross-sections, and comparisons of subgroups in each of between 12 demographic, comorbid, and clinical variables.

Methods

Two long-term glaucoma clinical trials used Humphrey Field Analyzer 24-2 VFs to calculate pointwise deviations from age-matched normal controls. Slopes of glaucomatous progression over time were calculated per participant using linear mixed models. Pointwise differences between subgroups in slopes and cross-sectional categories were tested, adjusting for multiple comparisons using false discovery rate (FDR) and Q values.

Results

Pointwise data were available for 1118 patients who had 15,073 VFs. On average, defects were seen at all VF points. Of the 12 variables, six had average pointwise slopes where Subgroup 1 had significantly faster progression than Subgroup 2 at all or many of the 52 VF points: participants who were older (≥65 vs. younger), 52/52; were male, 13/52; had diabetes, 29/52; had hypertension, 46/52; had a larger cup-to-disc ratio (≥0.7), 36/52; or had larger differences in absolute mean deviation (MD) between eyes (>3 dB), 52/52. Cross-sectional patterns at MD severity of -12 to -6.1 dB showed strong midline effects for gender and other patterns for hypertension, cup-to-disc ratio, absolute difference in MD between eyes, and disc notching.

Conclusions

The approach used provides new longitudinal and cross-sectional insights into variation across the VF associated with demographic, comorbid, and clinical variables.

Translational Relevance

This exploration and characterization of variable effects in the setting of pointwise VF testing may enable clinicians to anticipate patterns of VF loss based on demographic, comorbid, and clinical associations.

The Relationship Between Optic Disc and Retinal Artery Position and Glaucomatous Visual Field Progression

Purpose

To investigate whether retinal structural parameters, including positions of the optic disc and major retinal arteries, affect glaucomatous progression of the visual field (VF).

Methods

In this cohort study, 116 eyes of 73 patients with primary open angle glaucoma (POAG) were included. VFs were measured using the Humphrey Field Analyzer 24-2 program and the VF was divided into seven sectors according to the corresponding optic disc angle. Average total deviation (TD) was calculated in each sector. Positions of major retinal arteries in the superotemporal and inferotemporal areas were decided by identifying the points where the retinal artery intersected the 3.4-mm-diameter circle around the optic disc. The relationship between sectorial TD VF progression rate and eight variables (age, mean and standard deviation of intraocular pressure during the observation period, baseline sectorial TD value, papillomacular bundle tilt angle, and axial length, along with superior/inferior arterial angle) was investigated.

Results

The main outcome measures were the association between retinal structural parameters and glaucomatous progression of VF. The superior retinal artery angular position was positively associated with sectorial TD progression rates in two central sectors in the inferior hemifield, which suggests faster VF progression where superior retinal artery angles are narrow. Papillomacular bundle tilt was not associated with TD progression rate in any sector.

Conclusions

Progression of the inferior VF was associated with the superior retinal artery angular position in this study of POAG.

Unsupervised Machine Learning Identifies Quantifiable Patterns of Visual Field Loss in Idiopathic Intracranial Hypertension

Purpose

Archetypal analysis, a form of unsupervised machine learning, identifies archetypal patterns within a visual field (VF) dataset such that any VF is described as a weighted sum of its archetypes (ATs) and has been used to quantify VF defects in glaucoma. We applied archetypal analysis to VFs affected by nonglaucomatous optic neuropathy caused by idiopathic intracranial hypertension (IIH).

Methods

We created an AT model from 2862 VFs prospectively collected from 330 eyes in the IIH Treatment Trial (IIHTT). We compared baseline IIH AT patterns with their descriptive VF classifications from the IIHTT.

Results

The optimum IIH AT model yielded 14 ATs resembling VF patterns reported in the IIHTT. Baseline VFs contained four or fewer meaningful ATs in 147 (89%) of study eyes. AT2 (mild general VF depression pattern) demonstrated the greatest number of study eyes with meaningful AT weight at baseline (n = 114), followed by AT1 (n = 91). Other ATs captured patterns of blind spot enlargement, hemianopia, arcuate, nasal defects, and more nonspecific patterns of general VF depression. Of all ATs, AT1 (normal pattern) had the strongest correlation with mean deviation (r = 0.69, P < 0.001). For 65 of the 93 VFs with a dominant AT, this AT matched the expert classification.

Conclusions

Archetypal analysis identifies quantifiable, archetypal VF defects that resemble those commonly seen in IIH.

Translational Relevance

Archetypal analysis provides a quantitative, objective method of measuring and monitoring disease-specific regional VF defects in IIH.

Advances in perimetry for glaucoma

PURPOSE OF REVIEW

Perimetry remains important for the diagnosis and management of glaucoma despite advances in imaging technology. The purpose of this review is to describe advances in the acquisition and analysis of visual field data and highlight novel techniques for performing perimetry.

RECENT FINDINGS

Studies have focused on improving the detection of patients at highest risk of severe vision loss and the development of innovative testing strategies that allow for more frequent testing. Artificial intelligence has been utilized in research settings to improve detection and characterization of glaucomatous field damage. Furthermore, tablet-based strategies and virtual reality headsets show promise for glaucoma screening and remote monitoring of patients with glaucoma.

SUMMARY

New testing strategies and research findings have improved our ability to identify patients with both paracentral and mid-peripheral visual field progression. New strategies have the potential to make visual field testing more efficient, reliable and accessible for patients with glaucoma.

Artificial intelligence and complex statistical modeling in glaucoma diagnosis and management

PURPOSE OF REVIEW

The field of artificial intelligence has grown exponentially in recent years with new technology, methods, and applications emerging at a rapid rate. Many of these advancements have been used to improve the diagnosis and management of glaucoma. We aim to provide an overview of recent publications regarding the use of artificial intelligence to enhance the detection and treatment of glaucoma.

RECENT FINDINGS

Machine learning classifiers and deep learning algorithms have been developed to autonomously detect early structural and functional changes of glaucoma using different imaging and testing modalities such as fundus photography, optical coherence tomography, and standard automated perimetry. Artificial intelligence has also been used to further delineate structure-function correlation in glaucoma. Additional 'structure-structure' predictions have been successfully estimated. Other machine learning techniques utilizing complex statistical modeling have been used to detect glaucoma progression, as well as to predict future progression. Although not yet approved for clinical use, these artificial intelligence techniques have the potential to significantly improve glaucoma diagnosis and management.

SUMMARY

Rapidly emerging artificial intelligence algorithms have been used for the detection and management of glaucoma. These algorithms may aid the clinician in caring for patients with this complex disease. Further validation is required prior to employing these techniques widely in clinical practice.

Inter-Eye Association of Visual Field Defects in Glaucoma and Its Clinical Utility

Purpose

To investigate intereye associations of visual field (VF) defects.

Methods

We selected 24-2 VF pairs of both eyes from 63,604 patients tested on the same date with mean deviation (MD) ≥ −12 dB. VFs were decomposed into one normal and 15 defect patterns previously identified using archetypal analysis. VF pattern weighting coefficients were correlated between the worse and better eyes, as defined by MD. VF defect patterns (weighting coefficients > 10%) in the better eye were predicted from weighting coefficients of the worse eye by logistic regression models, which were evaluated by area under the receiver operating characteristic curve (AUC).

Results

Intereye correlations of archetypal VF patterns were strongest for the same defect pattern between fellow eyes. The AUCs for predicting the presence of 15 defect patterns in the better eye based on the worse eye ranged from 0.69 (superior nasal step) to 0.92 (near total loss). The AUC for predicting superior paracentral loss was 0.89. Superior paracentral loss in the better eye was positively correlated with coefficients of superior paracentral loss, central scotoma, superior altitudinal defect, nasal hemianopia, and inferior paracentral loss in the worse eye, and negatively correlated with coefficients of the normal VF, superior peripheral defect, concentric peripheral defect, and temporal wedge. The parameters are presented in the descending order of statistical significance.

Conclusions

VF patterns of the worse eye are predictive of VF defects in the better eye.

Translational Relevance

Our models can potentially assist clinicians to better interpret VF loss under measurement uncertainty.

Clinical application of ultra-widefield fundus autofluorescence

PURPOSE

To review the basic principles of ultra-widefield fundus autofluorescence (UWF-FAF) and discuss its clinical application for a variety of retinal and choroidal disorders.

METHODS

A systematic review of the PubMed database was performed using the search terms "ultra-widefield," "autofluorescence," "retinal disease" and "choroidal disease."

RESULTS

UWF-FAF imaging is a recently developed noninvasive retinal imaging modality with a wide imaging range that can locate peripheral fundus lesions that traditional fundus autofluorescence cannot. Multiple commercially available ultra-widefield imaging systems, including Heidelberg Spectralis and Optomap Ultra-Widefield systems, are available to the clinician. Imaging by UWF-FAF is more comprehensive; it can reflect the content and distribution of the predominant ocular fluorophore in retinal pigment epithelial cells and evaluate the metabolic status of RPE of various retinal and choroidal disorders.

CONCLUSION

UWF-FAF can detect abnormalities that traditional fundus autofluorescence cannot; therefore, it can be used to better elucidate disease pathogenesis, analyze genotype-phenotype correlations, diagnose and monitor disease.