Trained artificial neural network for glaucoma diagnosis using visual field data: a comparison with conventional algorithms

Statistical performance review on diagnosis of leukemia, glaucoma and diabetes mellitus using AI

The growth of artificial intelligence (AI) in the healthcare industry tremendously increases the patient outcomes by reshaping the way we diagnose, treat and monitor patients. AI-based innovation in healthcare include exploration of drugs, personalized medicine, clinical diagnosis investigations, robotic-assisted surgery, verified prescriptions, pregnancy care for women, radiology, and reviewed patient information analytics. However, prediction of AI-based solutions are depends mainly on the implementation of statistical algorithms and input data set. In this article, statistical performance review on various algorithms, Accuracy, Precision, Recall and F1-Score used to predict the diagnosis of leukemia, glaucoma, and diabetes mellitus is presented. Review on statistical algorithms' performance, used for individual disease diagnosis gives a complete picture of various research efforts during the last two decades. At the end of statistical review on each disease diagnosis, we have discussed our inferences that will give future directions for the new researchers on selection of AI statistical algorithm as well as the input data set.

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.

Evaluating machine learning classifiers for glaucoma referral decision support in primary care settings

Several artificial intelligence algorithms have been proposed to help diagnose glaucoma by analyzing the functional and/or structural changes in the eye. These algorithms require carefully curated datasets with access to ocular images. In the current study, we have modeled and evaluated classifiers to predict self-reported glaucoma using a single, easily obtained ocular feature (intraocular pressure (IOP)) and non-ocular features (age, gender, race, body mass index, systolic and diastolic blood pressure, and comorbidities). The classifiers were trained on publicly available data of 3015 subjects without a glaucoma diagnosis at the time of enrollment. 337 subjects subsequently self-reported a glaucoma diagnosis in a span of 1–12 years after enrollment. The classifiers were evaluated on the ability to identify these subjects by only using their features recorded at the time of enrollment. Support vector machine, logistic regression, and adaptive boosting performed similarly on the dataset with F1 scores of 0.31, 0.30, and 0.28, respectively. Logistic regression had the highest sensitivity at 60% with a specificity of 69%. Predictive classifiers using primarily non-ocular features have the potential to be used for identifying suspected glaucoma in non-eye care settings, including primary care. Further research into finding additional features that improve the performance of predictive classifiers is warranted.

Glaucoma diagnosis using multi-feature analysis and a deep learning technique

In this study, we aimed to facilitate the current diagnostic assessment of glaucoma by analyzing multiple features and introducing a new cross-sectional optic nerve head (ONH) feature from optical coherence tomography (OCT) images. The data (n = 100 for both glaucoma and control) were collected based on structural, functional, demographic and risk factors. The features were statistically analyzed, and the most significant four features were used to train machine learning (ML) algorithms. Two ML algorithms: deep learning (DL) and logistic regression (LR) were compared in terms of the classification accuracy for automated glaucoma detection. The performance of the ML models was evaluated on unseen test data, n = 55. An image segmentation pilot study was then performed on cross-sectional OCT scans. The ONH cup area was extracted, analyzed, and a new DL model was trained for glaucoma prediction. The DL model was estimated using five-fold cross-validation and compared with two pre-trained models. The DL model trained from the optimal features achieved significantly higher diagnostic performance (area under the receiver operating characteristic curve (AUC) 0.98 and accuracy of 97% on validation data and 96% on test data) compared to previous studies for automated glaucoma detection. The second DL model used in the pilot study also showed promising outcomes (AUC 0.99 and accuracy of 98.6%) to detect glaucoma compared to two pre-trained models. In combination, the result of the two studies strongly suggests the four features and the cross-sectional ONH cup area trained using deep learning have a great potential for use as an initial screening tool for glaucoma which will assist clinicians in making a precise decision.

Deep Learning Image Analysis of Optical Coherence Tomography Angiography Measured Vessel Density Improves Classification of Healthy and Glaucoma Eyes

PURPOSE

To compare convolutional neural network (CNN) analysis of en face vessel density images to gradient boosting classifier (GBC) analysis of instrument-provided, feature-based optical coherence tomography angiography (OCTA) vessel density measurements and OCT retinal nerve fiber layer (RNFL) thickness measurements for classifying healthy and glaucomatous eyes.

DESIGN

Comparison of diagnostic approaches.

METHODS

A total of 130 eyes of 80 healthy individuals and 275 eyes of 185 glaucoma patients with optic nerve head (ONH) OCTA and OCT imaging were included. Classification performance of a VGG16 CNN trained and tested on entire en face 4.5 × 4.5-mm radial peripapillary capillary OCTA ONH images was compared to the performance of separate GBC models trained and tested on standard OCTA and OCT measurements. Five-fold cross-validation was used to test predictions for CNNs and GBCs. Areas under the precision recall curves (AUPRC) were calculated to control for training/test set size imbalance and were compared.

RESULTS

Adjusted AUPRCs for GBC models were 0.89 (95% CI = 0.82, 0.92) for whole image vessel density GBC, 0.89 (0.83, 0.92) for whole image capillary density GBC, 0.91 (0.88, 0.93) for combined whole image vessel and whole image capillary density GBC, and 0.93 (0.91, 095) for RNFL thickness GBC. The adjusted AUPRC using CNN analysis of en face vessel density images was 0.97 (0.95, 0.99) resulting in significantly improved classification compared to GBC OCTA-based results and GBC OCT-based results (P ≤ 0.01 for all comparisons).

CONCLUSION

Deep learning en face image analysis improves on feature-based GBC models for classifying healthy and glaucoma eyes.

[Artificial intelligence and glaucoma: A literature review]

In recent years, research in artificial intelligence (AI) has experienced an unprecedented surge in the field of ophthalmology, in particular glaucoma. The diagnosis and follow-up of glaucoma is complex and relies on a body of clinical evidence and ancillary tests. This large amount of information from structural and functional testing of the optic nerve and macula makes glaucoma a particularly appropriate field for the application of AI. In this paper, we will review work using AI in the field of glaucoma, whether for screening, diagnosis or detection of progression. Many AI strategies have shown promising results for glaucoma detection using fundus photography, optical coherence tomography, or automated perimetry. The combination of these imaging modalities increases the performance of AI algorithms, with results comparable to those of humans. We will discuss potential applications as well as obstacles and limitations to the deployment and validation of such models. While there is no doubt that AI has the potential to revolutionize glaucoma management and screening, research in the coming years will need to address unavoidable questions regarding the clinical significance of such results and the explicability of the predictions.

Artificial Intelligence Algorithms to Diagnose Glaucoma and Detect Glaucoma Progression: Translation to Clinical Practice

Purpose

This concise review aims to explore the potential for the clinical implementation of artificial intelligence (AI) strategies for detecting glaucoma and monitoring glaucoma progression.

Methods

Nonsystematic literature review using the search combinations “Artificial Intelligence,” “Deep Learning,” “Machine Learning,” “Neural Networks,” “Bayesian Networks,” “Glaucoma Diagnosis,” and “Glaucoma Progression.” Information on sensitivity and specificity regarding glaucoma diagnosis and progression analysis as well as methodological details were extracted.

Results

Numerous AI strategies provide promising levels of specificity and sensitivity for structural (e.g. optical coherence tomography [OCT] imaging, fundus photography) and functional (visual field [VF] testing) test modalities used for the detection of glaucoma. Area under receiver operating curve (AROC) values of > 0.90 were achieved with every modality. Combining structural and functional inputs has been shown to even more improve the diagnostic ability. Regarding glaucoma progression, AI strategies can detect progression earlier than conventional methods or potentially from one single VF test.

Conclusions

AI algorithms applied to fundus photographs for screening purposes may provide good results using a simple and widely accessible test. However, for patients who are likely to have glaucoma more sophisticated methods should be used including data from OCT and perimetry. Outputs may serve as an adjunct to assist clinical decision making, whereas also enhancing the efficiency, productivity, and quality of the delivery of glaucoma care. Patients with diagnosed glaucoma may benefit from future algorithms to evaluate their risk of progression. Challenges are yet to be overcome, including the external validity of AI strategies, a move from a “black box” toward “explainable AI,” and likely regulatory hurdles. However, it is clear that AI can enhance the role of specialist clinicians and will inevitably shape the future of the delivery of glaucoma care to the next generation.

Translational Relevance

The promising levels of diagnostic accuracy reported by AI strategies across the modalities used in clinical practice for glaucoma detection can pave the way for the development of reliable models appropriate for their translation into clinical practice. Future incorporation of AI into healthcare models may help address the current limitations of access and timely management of patients with glaucoma across the world.

Gradient-Boosting Classifiers Combining Vessel Density and Tissue Thickness Measurements for Classifying Early to Moderate Glaucoma

PURPOSE

To compare gradient-boosting classifier (GBC) analysis of optical coherence tomography angiography (OCTA)-measured vessel density (VD) and OCT-measured tissue thickness to standard OCTA VD and OCT thickness parameters for classifying healthy eyes and eyes with early to moderate glaucoma.

DESIGN

Comparison of diagnostic tools.

METHODS

A total of 180 healthy eyes and 193 glaucomatous eyes with OCTA and OCT imaging of the macula and optic nerve head (ONH) were studied. Four GBCs were evaluated that combined 1) all macula VD and thickness measurements (Macula GBC), 2) all ONH VD and thickness measurements (ONH GBC), 3) all VD measurements from the macula and ONH (vessel density GBC), and 4) all thickness measurements from the macula and ONH (thickness GBC). ROC curve (AUROC) analyses compared the diagnostic accuracy of GBCs to that of standard instrument-provided parameters. A fifth GBC that combined all parameters (full GBC) also was investigated.

RESULTS

GBCs had better diagnostic accuracy than standard OCTA and OCT parameters with AUROCs ranging from 0.90 to 0.93 and 0.64 to 0.91, respectively. The full GBC (AUROC = 0.93) performed significantly better than the ONH GBC (AUROC = 0.91; P = .036) and the vessel density GBC (AUROC = 0.90; P = .010). All other GBCs performed similarly. The mean relative influence of each parameter included in the full GBC identified a combination of macular thickness and ONH VD measurements as the greatest contributors.

CONCLUSIONS

GBCs that combine OCTA and OCT macula and ONH measurements can improve diagnostic accuracy for glaucoma detection compared to most but not all instrument provided parameters.

Artificial intelligence and deep learning in glaucoma: Current state and future prospects

Over the past few years, there has been an unprecedented and tremendous excitement for artificial intelligence (AI) research in the field of Ophthalmology; this has naturally been translated to glaucoma-a progressive optic neuropathy characterized by retinal ganglion cell axon loss and associated visual field defects. In this review, we aim to discuss how AI may have a unique opportunity to tackle the many challenges faced in the glaucoma clinic. This is because glaucoma remains poorly understood with difficulties in providing early diagnosis and prognosis accurately and in a timely fashion. In the short term, AI could also become a game changer by paving the way for the first cost-effective glaucoma screening campaigns. While there are undeniable technical and clinical challenges ahead, and more so than for other ophthalmic disorders whereby AI is already booming, we strongly believe that glaucoma specialists should embrace AI as a companion to their practice. Finally, this review will also remind ourselves that glaucoma is a complex group of disorders with a multitude of physiological manifestations that cannot yet be observed clinically. AI in glaucoma is here to stay, but it will not be the only tool to solve glaucoma.

Glaucoma management in the era of artificial intelligence

Glaucoma is a result of irreversible damage to the retinal ganglion cells. While an early intervention could minimise the risk of vision loss in glaucoma, its asymptomatic nature makes it difficult to diagnose until a late stage. The diagnosis of glaucoma is a complicated and expensive effort that is heavily dependent on the experience and expertise of a clinician. The application of artificial intelligence (AI) algorithms in ophthalmology has improved our understanding of many retinal, macular, choroidal and corneal pathologies. With the advent of deep learning, a number of tools for the classification, segmentation and enhancement of ocular images have been developed. Over the years, several AI techniques have been proposed to help detect glaucoma by analysis of functional and/or structural evaluations of the eye. Moreover, the use of AI has also been explored to improve the reliability of ascribing disease prognosis. This review summarises the role of AI in the diagnosis and prognosis of glaucoma, discusses the advantages and challenges of using AI systems in clinics and predicts likely areas of future progress.

Specificity of various cluster criteria used for the detection of glaucomatous visual field abnormalities

Purpose

This study aimed to evaluate the specificity of commonly used cluster criteria for defining the presence of glaucomatous visual field abnormalities and the impact of variations in the criterion used.

Methods

This is an observational study including 607 eyes from 384 healthy participants, and 501 eyes of 345 participants with glaucoma, with at least two reliable 24–2 visual field tests. An abnormal visual field cluster was defined as the presence of ≥3 contiguous abnormal locations. Variations in this definition were evaluated and included (1) whether abnormalities were based on total deviation and/or pattern deviation values; (2) probability cut-off for defining an abnormal location; and (3) whether abnormalities were required to be repeatable (within the same hemifield or at the same locations) or not. These definitions were also compared against pattern standard deviation (PSD) values.

Results

False-positive rates of various cluster criteria ranged between 9% and 46% depending on the specific definitions used. Only definitions that required abnormalities to be repeatable at the same location achieved a false-positive rate of ≤6%. The various cluster criteria generally performed similarly or worse at detecting glaucoma eyes compared with the PSD values.

Conclusions

Commonly used visual field cluster criteria have high false-positive rates that vary widely depending on the definition used. These findings highlight the need to carefully consider the criteria used when designing and interpreting glaucoma clinical studies.

Trial registration number

NCT00221923.

Social Determinants of Association among Diabetes Mellitus, Visual Impairment and Hearing Loss in a Middle-Aged or Old Population: Artificial-Neural-Network Analysis of the Korean Longitudinal Study of Aging (2014⁻2016)

BACKGROUND

This study introduces a new framework based on an artificial neural network (ANN) for testing whether social determinants are major determinants of association among diabetes mellitus, visual impairment and hearing loss in a middle-aged or old population.

METHODS

The data came from the Korean Longitudinal Study of Aging (2014⁻2016), with 6120 participants aged 45 years or more. The association was divided into eight categories: one category for having no disease, three categories for having one, three categories for having two and one category for having three. Variable importance, the effect of a variable on model performance, was used to evaluate the hypothesis based on whether family support, socioeconomic status and social activity in Y2014 are among the top 10 determinants of the association in the year 2016 (Y2016).

RESULTS

Based on variable importance from the ANN, brothers/sisters cohabiting (0.0167), voluntary activity (0.0148), income (0.0125), family activity (0.0125), parents alive (0.0121), leisure activity (0.0095) and meeting with friends (0.0092) in Y2014 are the top-10 determinants of comorbidity in Y2016.

CONCLUSION

The findings of this study support the hypothesis, highlighting the importance of social determinants for the effective management of the comorbidities of the three diseases.

Portable Perimetry Using Eye-Tracking on a Tablet Computer-A Feasibility Assessment

Purpose

Visual field (VF) examination by standard automated perimetry (SAP) is an important method of clinical assessment. However, the complexity of the test, and its use of bulky, expensive equipment makes it impractical for case-finding. We propose and evaluate a new approach to paracentral VF assessment that combines an inexpensive eye-tracker with a portable tablet computer (“Eyecatcher”).

Methods

Twenty-four eyes from 12 glaucoma patients, and 12 eyes from six age-similar controls were examined. Participants were tested monocularly (once per eye), with both the novel Eyecatcher test and traditional SAP (HFA SITA standard 24-2). For Eyecatcher, the participant's task was to simply to look at a sequence of fixed-luminance dots, presented relative to the current point of fixation. Start and end fixations were used to determine locations where stimuli were seen/unseen, and to build a continuous map of sensitivity loss across a VF of approximately 20°.

Results

Eyecatcher was able to clearly separate patients from controls, and the results were consistent with those from traditional SAP. In particular, mean Eyecatcher scores were strongly correlated with mean deviation scores (r² = 0.64, P < 0.001), and there was good concordance between corresponding VF locations (∼84%). Participants reported that Eyecatcher was more enjoyable, easier to perform, and less tiring than SAP (all P < 0.001).

Conclusions

Portable perimetry using an inexpensive eye-tracker and a tablet computer is feasible, although possible means of improvement are suggested.

Translational Relevance

Such a test could have significant utility as a case finding device.

Artificial intelligence in glaucoma

PURPOSE OF REVIEW

The use of computers has become increasingly relevant to medical decision-making, and artificial intelligence methods have recently demonstrated significant advances in medicine. We therefore provide an overview of current artificial intelligence methods and their applications, to help the practicing ophthalmologist understand their potential impact on glaucoma care.

RECENT FINDINGS

Techniques used in artificial intelligence can successfully analyze and categorize data from visual fields, optic nerve structure [e.g., optical coherence tomography (OCT) and fundus photography], ocular biomechanical properties, and a combination thereof to identify disease severity, determine disease progression, and/or recommend referral for specialized care. Algorithms have become increasingly complex in recent years, utilizing both supervised and unsupervised methods of artificial intelligence. Impressive performance of these algorithms on previously unseen data has been reported, often outperforming standard global indices and expert observers. However, there remains no clearly defined gold standard for determining the presence and severity of glaucoma, which undermines the training of these algorithms. To improve upon existing methodologies, future work must employ more robust definitions of disease, optimize data inputs for artificial intelligence analysis, and improve methods of extracting knowledge from learned results.

SUMMARY

Artificial intelligence has the potential to revolutionize the screening, diagnosis, and classification of glaucoma, both through the automated processing of large data sets, and by earlier detection of new disease patterns. In addition, artificial intelligence holds promise for fundamentally changing research aimed at understanding the development, progression, and treatment of glaucoma, by identifying novel risk factors and by evaluating the importance of existing ones.

Evaluation of Pupil Fields Using a Newly Developed Perimeter in Glaucoma Patients

PURPOSE

To evaluate objective pupil fields using a newly developed perimeter for the detection of glaucomatous damage.

MATERIALS AND METHODS

Forty-three eyes of 32 glaucoma patients (42-69 years) were examined. Glaucomatous eyes were classified into three stages using the Hodapp-Anderson-Parrish grading scale (early, 16; moderate, 14; and severe, 13 eyes). The head-mounted perimeter "imo" was used to measure the percentage pupil constriction (PPC) of the pupil fields at 36 test points. A stimulus target size of Goldmann V with 0 decibels (dB) light under 31.4 apostilbs (asb) background was presented. Visual fields were measured with the Humphrey Field Analyzer 10-2 program. Using the 3D OCT-2000, 10 × 10 grid of the macular thickness were also obtained. Median correlation coefficients (r) of each examined eye were analyzed between the PPC and visual field sensitivity (dB), and the thickness of the retinal nerve fiber layer (RNFL), ganglion cell layer (GCL)⁺ (GCL + inner plexiform layer [IPL]), and GCL⁺⁺ (RNFL + GCL + IPL), respectively.

RESULTS

Moderate correlations between the PPC and dB (r = 0.44-0.55), and GCL⁺⁺ (r = 0.43-0.45) were obtained in the correspondence analysis of 12 test points. There were no significant differences in glaucoma severity (P = 0.924-1.000). However, some patients with extremely early stage glaucoma (visual field index ≥90%) tended to have poor correlation.

CONCLUSIONS

Pupil fields of the imo generally corresponded to the visual fields and the RNFL + GCL + IPL thickness, even in early glaucoma; however, the examiner must clearly understand the criteria of patient selection.

Performance of the 10-2 and 24-2 Visual Field Tests for Detecting Central Visual Field Abnormalities in Glaucoma

PURPOSE

To compare the performance of the pattern standard deviation (PSD) values derived from the central 12 locations of the 24-2 visual field test (C24-2) to the entire 10-2 test for detecting central visual field abnormalities in eyes with, suspected of having, or at risk of having glaucoma.

DESIGN

Cross-sectional case-control study.

METHODS

Eyes with, suspected of having, or at risk of having glaucoma, based on masked grading of optic disc stereophotographs and/or ocular hypertension (intraocular pressure ≥ 22 mm Hg) were included as cases (n = 523). Eyes from healthy participants were included as controls (n = 107) to allow the 2 tests to be compared at matched specificities. The sensitivity to detect cases at 95% specificity using PSD values derived from the entire 10-2 test and C24-2 were compared.

RESULTS

The sensitivity of the 10-2 and C24-2 PSD values was not significantly different between the 10-2 and C24-2 at matched specificities (35.9% and 35.4% respectively; P = .900). There was also a substantial agreement between the cases detected by both methods (kappa = 0.80 ± 0.04), and a very strong association between the PSD values from the 2 methods (R² = 0.91).

CONCLUSIONS

10-2 and 24-2 tests identified a similar number of eyes with, suspected of having, or at risk of having glaucoma as having central visual field abnormalities using PSD values. These findings do not mean that 10-2 tests are not useful, but highlight the need for further studies to determine the potential advantages of 10-2 tests through equivalent comparisons against 24-2 tests to ensure appropriate recommendations are made about its incorporation into the glaucoma standard of care.

A deep learning approach to automatic detection of early glaucoma from visual fields

PURPOSE

To investigate the suitability of multi-scale spatial information in 30o visual fields (VF), computed from a Convolutional Neural Network (CNN) classifier, for early-glaucoma vs. control discrimination.

METHOD

Two data sets of VFs acquired with the OCTOPUS 101 G1 program and the Humphrey Field Analyzer 24-2 pattern were subdivided into control and early-glaucomatous groups, and converted into a new image using a novel voronoi representation to train a custom-designed CNN so to discriminate between control and early-glaucomatous eyes. Saliency maps that highlight what regions of the VF are contributing maximally to the classification decision were computed to provide classification justification. Model fitting was cross-validated and average precision (AP) score performances were computed for our method, Mean Defect (MD), square-root of Loss Variance (sLV), their combination (MD+sLV), and a Neural Network (NN) that does not use convolutional features.

RESULTS

CNN achieved the best AP score (0.874±0.095) across all test folds for one data set compared to others (MD = 0.869±0.064, sLV = 0.775±0.137, MD+sLV = 0.839±0.085, NN = 0.843±0.089) and the third best AP score (0.986 ±0.019) on the other one with slight difference from the other methods (MD = 0.986±0.023, sLV = 0.992±0.016, MD+sLV = 0.987±0.017, NN = 0.985±0.017). In general, CNN consistently led to high AP across different data sets. Qualitatively, computed saliency maps appeared to provide clinically relevant information on the CNN decision for individual VFs.

CONCLUSION

The proposed CNN offers high classification performance for the discrimination of control and early-glaucoma VFs when compared with standard clinical decision measures. The CNN classification, aided by saliency visualization, may support clinicians in the automatic discrimination of early-glaucomatous and normal VFs.

New Technologies for Glaucoma Detection

This review describes some of the most recent advances in the development and application of new technologies for detecting and managing glaucoma, including imaging, visual function testing, and tonometry. The widespread availability of mobile technology in the developing world is improving health care delivery, for example, with smartphones and mobile applications that allow patient data to be assessed remotely by health care providers.

Unsupervised Gaussian Mixture-Model With Expectation Maximization for Detecting Glaucomatous Progression in Standard Automated Perimetry Visual Fields

Purpose

To validate Gaussian mixture-model with expectation maximization (GEM) and variational Bayesian independent component analysis mixture-models (VIM) for detecting glaucomatous progression along visual field (VF) defect patterns (GEM–progression of patterns (POP) and VIM-POP). To compare GEM-POP and VIM-POP with other methods.

Methods

GEM and VIM models separated cross-sectional abnormal VFs from 859 eyes and normal VFs from 1117 eyes into abnormal and normal clusters. Clusters were decomposed into independent axes. The confidence limit (CL) of stability was established for each axis with a set of 84 stable eyes. Sensitivity for detecting progression was assessed in a sample of 83 eyes with known progressive glaucomatous optic neuropathy (PGON). Eyes were classified as progressed if any defect pattern progressed beyond the CL of stability. Performance of GEM-POP and VIM-POP was compared to point-wise linear regression (PLR), permutation analysis of PLR (PoPLR), and linear regression (LR) of mean deviation (MD), and visual field index (VFI).

Results

Sensitivity and specificity for detecting glaucomatous VFs were 89.9% and 93.8%, respectively, for GEM and 93.0% and 97.0%, respectively, for VIM. Receiver operating characteristic (ROC) curve areas for classifying progressed eyes were 0.82 for VIM-POP, 0.86 for GEM-POP, 0.81 for PoPLR, 0.69 for LR of MD, and 0.76 for LR of VFI.

Conclusions

GEM-POP was significantly more sensitive to PGON than PoPLR and linear regression of MD and VFI in our sample, while providing localized progression information.

Translational Relevance

Detection of glaucomatous progression can be improved by assessing longitudinal changes in localized patterns of glaucomatous defect identified by unsupervised machine learning.

Detecting glaucomatous change in visual fields: Analysis with an optimization framework

Detecting glaucomatous progression is an important aspect of glaucoma management. The assessment of longitudinal series of visual fields, measured using Standard Automated Perimetry (SAP), is considered the reference standard for this effort. We seek efficient techniques for determining progression from longitudinal visual fields by formulating the problem as an optimization framework, learned from a population of glaucoma data. The longitudinal data from each patient's eye were used in a convex optimization framework to find a vector that is representative of the progression direction of the sample population, as a whole. Post-hoc analysis of longitudinal visual fields across the derived vector led to optimal progression (change) detection. The proposed method was compared to recently described progression detection methods and to linear regression of instrument-defined global indices, and showed slightly higher sensitivities at the highest specificities than other methods (a clinically desirable result). The proposed approach is simpler, faster, and more efficient for detecting glaucomatous changes, compared to our previously proposed machine learning-based methods, although it provides somewhat less information. This approach has potential application in glaucoma clinics for patient monitoring and in research centers for classification of study participants.

Exploring early glaucoma and the visual field test: classification and clustering using Bayesian networks

Bayesian networks (BNs) are probabilistic models used for classification and clustering in several fields. Their ability to deal with unobserved variables and to integrate data and expert knowledge make them an appropriate technique for modeling eye functionality measurements in glaucoma. In this study, a set of BNs is used to simultaneously perform classification of early glaucoma and cluster data into different stages of disease. A novel learning algorithm that combines clustering and quasi-greedy search is also proposed. The classification performances of the models are evaluated on an independent dataset, while the clusters are compared to K-means, previous publications, and direct knowledge. The use of clustering and structure learning enabled the exploration of the visual field patterns of the disease while obtaining good results both on pre- (50% sensitivity at 90% specificity) and post- (85% sensitivity at 90% specificity) diagnosis data. Clusters obtained were insightful and in conformity with consolidated knowledge in the field.

Development and validation of an improved neurological hemifield test to identify chiasmal and postchiasmal lesions by automated perimetry

PURPOSE

To improve the neurological hemifield test (NHT) using visual field data from both eyes to detect and classify visual field loss caused by chiasmal or postchiasmal lesions.

METHODS

Visual field and clinical data for 633 patients were divided into a training set (474 cases) and a validation set (159 cases). Each set had equal numbers of neurological, glaucoma, or glaucoma suspect cases, matched for age and for mean deviation between neurological and glaucoma cases. NHT scores as previously described and a new NHT laterality score were calculated. The ability of these scores to distinguish neurological from other fields was assessed with receiver operating characteristic (ROC) analysis. Three machine classifier algorithms were also evaluated: decision tree, random forest, and least absolute shrinkage and selection operator (LASSO). We also evaluated the ability of NHT to identify the type of neurological field defect (homonymous or bitemporal).

RESULTS

The area under the ROC curve (AUC) for the maximum NHT score was 0.92 (confidence interval [CI]: 0.87, 0.97). Using NHT laterality scores from each eye combined with the sum of NHT scores, the AUC improved to 0.93 (CI: 0.88, 0.98). The largest AUC for machine learning algorithms was for the LASSO method (0.96, CI: 0.92, 0.99). The NHT scores identified the type of neurological defect in 96% (158/164) of patients.

CONCLUSIONS

The new NHT distinguished neurological field defects from those of glaucoma and glaucoma suspects, providing accurate categorization of defect type. Its implementation may identify unsuspected neurological disease in clinical visual field testing.

Automatic segmentation of up to ten layer boundaries in SD-OCT images of the mouse retina with and without missing layers due to pathology

Accurate quantification of retinal layer thicknesses in mice as seen on optical coherence tomography (OCT) is crucial for the study of numerous ocular and neurological diseases. However, manual segmentation is time-consuming and subjective. Previous attempts to automate this process were limited to high-quality scans from mice with no missing layers or visible pathology. This paper presents an automatic approach for segmenting retinal layers in spectral domain OCT images using sparsity based denoising, support vector machines, graph theory, and dynamic programming (S-GTDP). Results show that this method accurately segments all present retinal layer boundaries, which can range from seven to ten, in wild-type and rhodopsin knockout mice as compared to manual segmentation and has a more accurate performance as compared to the commercial automated Diver segmentation software.

Glaucomatous patterns in Frequency Doubling Technology (FDT) perimetry data identified by unsupervised machine learning classifiers

Purpose

The variational Bayesian independent component analysis-mixture model (VIM), an unsupervised machine-learning classifier, was used to automatically separate Matrix Frequency Doubling Technology (FDT) perimetry data into clusters of healthy and glaucomatous eyes, and to identify axes representing statistically independent patterns of defect in the glaucoma clusters.

Methods

FDT measurements were obtained from 1,190 eyes with normal FDT results and 786 eyes with abnormal FDT results from the UCSD-based Diagnostic Innovations in Glaucoma Study (DIGS) and African Descent and Glaucoma Evaluation Study (ADAGES). For all eyes, VIM input was 52 threshold test points from the 24-2 test pattern, plus age.

Results

FDT mean deviation was −1.00 dB (S.D. = 2.80 dB) and −5.57 dB (S.D. = 5.09 dB) in FDT-normal eyes and FDT-abnormal eyes, respectively (p<0.001). VIM identified meaningful clusters of FDT data and positioned a set of statistically independent axes through the mean of each cluster. The optimal VIM model separated the FDT fields into 3 clusters. Cluster N contained primarily normal fields (1109/1190, specificity 93.1%) and clusters G₁ and G₂ combined, contained primarily abnormal fields (651/786, sensitivity 82.8%). For clusters G₁ and G₂ the optimal number of axes were 2 and 5, respectively. Patterns automatically generated along axes within the glaucoma clusters were similar to those known to be indicative of glaucoma. Fields located farther from the normal mean on each glaucoma axis showed increasing field defect severity.

Conclusions

VIM successfully separated FDT fields from healthy and glaucoma eyes without a priori information about class membership, and identified familiar glaucomatous patterns of loss.

Glaucoma Diagnostic Accuracy of Machine Learning Classifiers Using Retinal Nerve Fiber Layer and Optic Nerve Data from SD-OCT

Purpose. To investigate the diagnostic accuracy of machine learning classifiers (MLCs) using retinal nerve fiber layer (RNFL) and optic nerve (ON) parameters obtained with spectral domain optical coherence tomography (SD-OCT). Methods. Fifty-seven patients with early to moderate primary open angle glaucoma and 46 healthy patients were recruited. All 103 patients underwent a complete ophthalmological examination, achromatic standard automated perimetry, and imaging with SD-OCT. Receiver operating characteristic (ROC) curves were built for RNFL and ON parameters. Ten MLCs were tested. Areas under ROC curves (aROCs) obtained for each SD-OCT parameter and MLC were compared. Results. The mean age was 56.5 ± 8.9 years for healthy individuals and 59.9 ± 9.0 years for glaucoma patients (P = 0.054). Mean deviation values were −1.4 dB for healthy individuals and −4.0 dB for glaucoma patients (P < 0.001). SD-OCT parameters with the greatest aROCs were cup/disc area ratio (0.846) and average cup/disc (0.843). aROCs obtained with classifiers varied from 0.687 (CTREE) to 0.877 (RAN). The aROC obtained with RAN (0.877) was not significantly different from the aROC obtained with the best single SD-OCT parameter (0.846) (P = 0.542). Conclusion. MLCs showed good accuracy but did not improve the sensitivity and specificity of SD-OCT for the diagnosis of glaucoma.

Comparison of clinicians and an artificial neural network regarding accuracy and certainty in performance of visual field assessment for the diagnosis of glaucoma

PURPOSE

To compare clinicians and a trained artificial neural network (ANN) regarding accuracy and certainty of assessment of visual fields for the diagnosis of glaucoma.

METHODS

Thirty physicians with different levels of knowledge and experience in glaucoma management assessed 30-2 SITA Standard visual field printouts that included full Statpac information from 99 patients with glaucomatous optic neuropathy and 66 healthy subjects. Glaucomatous eyes with perimetric mean deviation values worsethan -10 dB were not eligible. The fields were graded on a scale of 1-10, where 1 indicated healthy with absolute certaintyand 10 signified glaucoma; 5.5 was the cut-off between healthy and glaucoma. The same fields were classified by a previously trained ANN. The ANN output was transformed into a linear scale that matched the scale used in the subjective assessments. Classification certainty was assessed using a classification error score.

RESULTS

Among the physicians, sensitivity ranged from 61% to 96% (mean 83%) and specificity from 59% to 100% (mean 90%). Our ANN achieved 93% sensitivity and 91% specificity, and it was significantly more sensitive than the physicians (p < 0.001) at a similar level of specificity. The ANN classification error score was equivalent to the top third scores of all physicians, and the ANN never indicated a high degree of certainty for any of its misclassified visual field tests.

CONCLUSION

Our results indicate that a trained ANN performs at least as well as physicians in assessments of visual fields for the diagnosis of glaucoma.

Glaucoma diagnostics

This thesis addresses several aspects of glaucoma diagnostics from both a clinical and a screening perspective. New instruments for diagnosing glaucoma have been developed over the past years, but little information is available regarding their performance as screening methods and their usefulness in ordinary clinical practice. PURPOSE OF THE RESEARCH UNDERLYING THIS THESIS:  The objectives of this research were as follows: to compare the accuracy of results of analysis of the optic nerve head (ONH) achieved by computerized imaging using the Heidelberg Retina Tomograph (HRT) and by subjective assessment performed by physicians with different degrees of experience of glaucoma (paper III); to evaluate the effect of a continuous medical education (CME) lecture on subjective assessment of the ONH for diagnosis of glaucoma (paper II); to investigate subjective assessment of perimetric test results by physicians with varying knowledge of glaucoma with a trained artificial neural network (ANN) and to compare the certainty of the classifications (paper IV); and to compare the diagnostic performance of time-domain Stratus optical coherence tomography (OCT) with that of spectral-domain Cirrus OCT (paper I), frequency doubling technology (FDT) screening perimetry and scanning laser polarimetry with the GDx variable corneal compensator (VCC) in a random population-based sample and in patients with glaucoma of varying disease severity.

METHODS AND RESULTS

  In evaluation of the ONH, use of the HRT statistical tools, Moorfields regression analysis (MRA) and the Glaucoma Probability Score (GPS) was compared with subjective assessment performed by 45 physicians. Optic nerve head images and photographs from 138 healthy and 97 glaucoma subjects were included. The sensitivity of MRA was higher (87-94%) than that of the average physician (62-82%), considerably greater than that of ophthalmologists with subspecialties other than glaucoma (53-77%) and non-significantly better than that of glaucoma experts (72-88%). Sensitivity achieved by GPS (79-93%) was also greater than that of the average physician. MRA correctly classified all eyes with advanced glaucomatous visual field defects, a result that was not achieved by GPS or even by the glaucoma experts. In eyes with small discs, MRA sensitivity (88%) was comparable with that of glaucoma experts (85%) and much better than that of GPS (50%). Also, the group comprising all physicians provided specificity (75-92%) similar to that of both MRA (69 - 86%) and GPS (72-94%) (Andersson et al. 2011a). A 1-hr CME lecture on ONH assessment led to a significant improvement in sensitivity (from 70% to 80%) and a significant decrease in uncertain assessments (from 22% to 13%), whereas specificity remained unchanged (68%) (Andersson et al. 2011b). A rise in sensitivity was seen in all subgroups of physicians, including glaucoma experts. Thirty physicians assessing standard automated perimetry (SAP) test results as Humphrey Field Analyzer single-field analysis printouts with full StatPac information from 99 patients with glaucoma and 66 healthy subjects were compared with a trained ANN regarding diagnostic performance. ANN reached significantly higher sensitivity (93%) than the average physician (83%), whereas specificity was similar for these two groups (91% and 90%, respectively). Diagnostic accuracy was similar among the different groups of physicians and seemingly rather independent of experience. Sensitivity ranged from 82% in the subgroup of other subspecialists to 87% in the glaucoma expert group, and specificity ranged from 88% among general ophthalmologists to 91% for glaucoma experts. The ANN attained certainty of classification that was in parity with that provided by the glaucoma experts and did not make any completely incorrect classifications of the visual fields (i.e. erroneous classifications were in the borderline zone) (Andersson et al. 2012). From a population-based randomly selected sample (n=308) of older subjects (aged ≥ 50 years) living in southern Sweden, 170 subjects underwent a comprehensive examination that included Stratus OCT, Cirrus OCT, an FDT screening programme and the GDx VCC. The same test protocol was applied to 138 randomized clinical patients with different stages of glaucoma. In the population-based sample, both Stratus and Cirrus OCT showed high diagnostic accuracy with area under the receiver-operating curve (aROC) values close to 1.0 (Bengtsson et al. 2012). Both OCT instruments correctly classified all of the clinical glaucoma patients with advanced disease. FDT screening showed high sensitivity (91%) but erroneously gave normal test results for some eyes with advanced disease. GDx VCC had lower sensitivity (73-92%) and also led to a large proportion of examinations with an atypical retardation pattern that is known to affect the diagnostic efficiency of this instrument.

CONCLUSIONS

  The HRT MRA performed better than most physicians and was consistent with the glaucoma experts. These results suggest that MRA can be a valuable tool for diagnosing glaucoma in ordinary practice, particularly when only a few glaucoma experts are available. Even though MRA provided 100% sensitivity in eyes with advanced glaucoma, it probably does not offer sufficient specificity to make it suitable as a screening method. Continuing medical education on ONH analysis had a small, but positive effect on diagnostic accuracy for glaucoma. An ANN trained to classify visual fields seemed to perform at least as well as most of the participating physicians, whose performances were remarkably similar regardless of their level of experience. This indicates that available tools for interpreting SAP findings are helpful in assessments of visual field test results. However, SAP is associated with learning effects (Heijl et al. 1989) that may entail low specificity for untrained subjects, and hence, it is not an ideal screening method for glaucoma. By comparison, the screening test of FDT is rapid and easy, but it is probably less suitable for screening purpose, because some eyes with advanced glaucoma were missed in this investigation. GDx VCC images for a relatively large number of eyes could not be analysed and is thus not appropriate for screening. The OCT instruments offer both high sensitivity and high specificity, and all eyes with advanced disease were correctly classified as glaucomatous in this evaluation. However, these instruments are still expensive and require special operator skills. Additional development to obtain OCT instrument that is more compact, easier to use and less expensive might render such tomography suitable as a screening tool for glaucoma.

AUTOMATED GLAUCOMA DETECTION USING HYBRID FEATURE EXTRACTION IN RETINAL FUNDUS IMAGES

Glaucoma is one of the most common causes of blindness. Robust mass screening may help to extend the symptom-free life for affected patients. To realize mass screening requires a cost-effective glaucoma detection method which integrates well with digital medical and administrative processes. To address these requirements, we propose a novel low cost automated glaucoma diagnosis system based on hybrid feature extraction from digital fundus images. The paper discusses a system for the automated identification of normal and glaucoma classes using higher order spectra (HOS), trace transform (TT), and discrete wavelet transform (DWT) features. The extracted features are fed to a support vector machine (SVM) classifier with linear, polynomial order 1, 2, 3 and radial basis function (RBF) in order to select the best kernel for automated decision making. In this work, the SVM classifier, with a polynomial order 2 kernel function, was able to identify glaucoma and normal images with an accuracy of 91.67%, and sensitivity and specificity of 90% and 93.33%, respectively. Furthermore, we propose a novel integrated index called Glaucoma Risk Index (GRI) which is composed from HOS, TT, and DWT features, to diagnose the unknown class using a single feature. We hope that this GRI will aid clinicians to make a faster glaucoma diagnosis during the mass screening of normal/glaucoma images.

Sensitivity and specificity of machine learning classifiers and spectral domain OCT for the diagnosis of glaucoma

Purpose. To investigate the sensitivity and specificity of machine learning classifiers (MLC) and spectral domain optical coherence tomography (SD-OCT) for the diagnosis of glaucoma. Methods. Sixty-two patients with early to moderate glaucomatous visual field damage and 48 healthy individuals were included. All subjects underwent a complete ophthalmologic examination, achromatic standard automated perimetry, and RNFL imaging with SD-OCT (Cirrus HD-OCT; Carl Zeiss Meditec, Inc., Dublin, California, USA). Receiver operating characteristic (ROC) curves were obtained for all SD-OCT parameters. Subsequently, the following MLCs were tested: Classification Tree (CTREE), Random Forest (RAN), Bagging (BAG), AdaBoost M1 (ADA), Ensemble Selection (ENS), Multilayer Perceptron (MLP), Radial Basis Function (RBF), Naive-Bayes (NB), and Support Vector Machine (SVM). Areas under the ROC curves (aROCs) obtained for each parameter and each MLC were compared. Results. The mean age was 57.0±9.2 years for healthy individuals and 59.9±9.0 years for glaucoma patients (p=0.103). Mean deviation values were -4.1±2.4 dB for glaucoma patients and -1.5±1.6 dB for healthy individuals (p<0.001). The SD-OCT parameters with the greater aROCs were inferior quadrant (0.813), average thickness (0.807), 7 o'clock position (0.765), and 6 o'clock position (0.754). The aROCs from classifiers varied from 0.785 (ADA) to 0.818 (BAG). The aROC obtained with BAG was not significantly different from the aROC obtained with the best single SD-OCT parameter (p=0.93). Conclusions. The SD-OCT showed good diagnostic accuracy in a group of patients with early glaucoma. In this series, MLCs did not improve the sensitivity and specificity of SD-OCT for the diagnosis of glaucoma.

Predicting glaucomatous progression in glaucoma suspect eyes using relevance vector machine classifiers for combined structural and functional measurements

PURPOSE

The goal of this study was to determine if glaucomatous progression in suspect eyes can be predicted from baseline confocal scanning laser ophthalmoscope (CSLO) and standard automated perimetry (SAP) measurements analyzed with relevance vector machine (RVM) classifiers.

METHODS

Two hundred sixty-four eyes of 193 participants were included. All eyes had normal SAP results at baseline with five or more SAP tests over time. Eyes were labeled progressed (n = 47) or stable (n = 217) during follow-up based on SAP Guided Progression Analysis or serial stereophotograph assessment. Baseline CSLO-measured topographic parameters (n = 117) and baseline total deviation values from the 24-2 SAP test-grid (n = 52) were selected from each eye. Ten-fold cross-validation was used to train and test RVMs using the CSLO and SAP features. Receiver operating characteristic (ROC) curve areas were calculated using full and optimized feature sets. ROC curve results from RVM analyses of CSLO, SAP, and CSLO and SAP combined were compared to CSLO and SAP global indices (Glaucoma Probability Score, mean deviation and pattern standard deviation).

RESULTS

The areas under the ROC curves (AUROCs) for RVMs trained on optimized feature sets of CSLO parameters, SAP parameters, and CSLO and SAP parameters combined were 0.640, 0.762, and 0.805, respectively. AUROCs for CSLO Glaucoma Probability Score, SAP mean deviation (MD), and SAP pattern standard deviation (PSD) were 0.517, 0.513, and 0.620, respectively. No CSLO or SAP global indices discriminated between baseline measurements from progressed and stable eyes better than chance.

CONCLUSIONS

In our sample, RVM analyses of baseline CSLO and SAP measurements could identify eyes that showed future glaucomatous progression with a higher accuracy than the CSLO and SAP global indices. (ClinicalTrials.gov numbers, NCT00221897, NCT00221923.).

What is causing the corneal ulcer? Management strategies for unresponsive corneal ulceration

Infectious keratitis represents a significant cause of ocular morbidity in the United States. The work-up and treatment of presumed infectious keratitis (PIK) has changed in the past two decades. The development of newer topical antibiotics has enabled broad-spectrum antibiotic coverage with good tissue penetration. The majority of PIK cases respond well to this strategy. The small numbers of cases that do not respond to the treatment are the cases that offer a diagnostic and therapeutic challenge. This review will describe different algorithms that can be followed for the successful management of patients with difficult or progressive PIK. These algorithms are based on scientific work and on our empirical clinical experience. The review will also present three different clinical cases of PIK that were managed according to the algorithms presented in this review.

Evaluation of an algorithm for detecting visual field defects due to chiasmal and postchiasmal lesions: the neurological hemifield test

PURPOSE

To develop an automated neurologic hemifield test (NHT) to detect visual field loss caused by chiasmal or postchiasmal lesions.

METHODS

Visual field locations from 24-2 pattern automated visual fields were grouped into two symmetric regions with 16 points on either side of the vertical meridian. A scoring system similar to the Glaucoma Hemifield Test (GHT) was used to calculate point scores using the pattern deviation values from the right and left regions. The cross-vertical difference in the sum of these values was the NHT score. The NHT was evaluated using visual fields from subjects with known neurologic disease, subjects with glaucoma, and glaucoma suspects (92 pairs of eyes each). The NHT score was calculated for each eye. Four masked reviewers scored all pairs of visual fields with regard to the likelihood of neurologic and glaucomatous optic neuropathy. Both NHT score and expert field ratings were compared with clinical diagnosis by receiver operating characteristic (ROC) analysis.

RESULTS

The NHT effectively discriminated neurologic fields from those of glaucoma patients and glaucoma suspects (area under the ROC curve [AUC] = 0.90; 95% confidence interval [CI], 0.86-0.94). The NHT score correlated well with clinician grading (Pearson correlation estimates, 0.74-0.78). Even when field defects were subtle, the NHT had some ability to discriminate neurologic from nonneurologic fields (AUC 0.68; 95% CI, 0.56-0.79).

CONCLUSIONS

The NHT distinguished neurologic field defects from those of glaucoma and glaucoma suspects, rivaling the performance of subspecialist clinicians. Its implementation may help identify unsuspected neurologic disease.

Integration and fusion of standard automated perimetry and optical coherence tomography data for improved automated glaucoma diagnostics

Background

The performance of glaucoma diagnostic systems could be conceivably improved by the integration of functional and structural test measurements that provide relevant and complementary information for reaching a diagnosis. The purpose of this study was to investigate the performance of data fusion methods and techniques for simple combination of Standard Automated Perimetry (SAP) and Optical Coherence Tomography (OCT) data for the diagnosis of glaucoma using Artificial Neural Networks (ANNs).

Methods

Humphrey 24-2 SITA standard SAP and StratusOCT tests were prospectively collected from a randomly selected population of 125 healthy persons and 135 patients with glaucomatous optic nerve heads and used as input for the ANNs. We tested commercially available standard parameters as well as novel ones (fused OCT and SAP data) that exploit the spatial relationship between visual field areas and sectors of the OCT peripapillary scan circle. We evaluated the performance of these SAP and OCT derived parameters both separately and in combination.

Results

The diagnostic accuracy from a combination of fused SAP and OCT data (95.39%) was higher than that of the best conventional parameters of either instrument, i.e. SAP Glaucoma Hemifield Test (p < 0.001) and OCT Retinal Nerve Fiber Layer Thickness ≥ 1 quadrant (p = 0.031). Fused OCT and combined fused OCT and SAP data provided similar Area under the Receiver Operating Characteristic Curve (AROC) values of 0.978 that were significantly larger (p = 0.047) compared to ANNs using SAP parameters alone (AROC = 0.945). On the other hand, ANNs based on the OCT parameters (AROC = 0.970) did not perform significantly worse than the ANNs based on the fused or combined forms of input data. The use of fused input increased the number of tests that were correctly classified by both SAP and OCT based ANNs.

Conclusions

Compared to the use of SAP parameters, input from the combination of fused OCT and SAP parameters, and from fused OCT data, significantly increased the performance of ANNs. Integrating parameters by including a priori relevant information through data fusion may improve ANN classification accuracy compared to currently available methods.

Machine learning classifiers for glaucoma diagnosis based on classification of retinal nerve fibre layer thickness parameters measured by Stratus OCT

PURPOSE

To compare the performance of two machine learning classifiers (MLCs), artificial neural networks (ANNs) and support vector machines (SVMs), with input based on retinal nerve fibre layer thickness (RNFLT) measurements by optical coherence tomography (OCT), on the diagnosis of glaucoma, and to assess the effects of different input parameters.

METHODS

We analysed Stratus OCT data from 90 healthy persons and 62 glaucoma patients. Performance of MLCs was compared using conventional OCT RNFLT parameters plus novel parameters such as minimum RNFLT values, 10th and 90th percentiles of measured RNFLT, and transformations of A-scan measurements. For each input parameter and MLC, the area under the receiver operating characteristic curve (AROC) was calculated.

RESULTS

There were no statistically significant differences between ANNs and SVMs. The best AROCs for both ANN (0.982, 95%CI: 0.966-0.999) and SVM (0.989, 95% CI: 0.979-1.0) were based on input of transformed A-scan measurements. Our SVM trained on this input performed better than ANNs or SVMs trained on any of the single RNFLT parameters (p < or = 0.038). The performance of ANNs and SVMs trained on minimum thickness values and the 10th and 90th percentiles were at least as good as ANNs and SVMs with input based on the conventional RNFLT parameters.

CONCLUSION

No differences between ANN and SVM were observed in this study. Both MLCs performed very well, with similar diagnostic performance. Input parameters have a larger impact on diagnostic performance than the type of machine classifier. Our results suggest that parameters based on transformed A-scan thickness measurements of the RNFL processed by machine classifiers can improve OCT-based glaucoma diagnosis.

The scintillating grid illusion: influence of size, shape, and orientation of the luminance patches

The scintillating grid illusion refers to the illusory perception of black spots on luminance patches at the intersections of a grey grid on a black background. We examined how spatial parameters of luminance patches modulated the strength of the illusion. In experiment 1, we controlled the size and shape of the luminance patches. For the largest-size conditions tested, we found a significant reduction in the strength of the illusion with squares when compared to circles or diamonds. In experiment 2, we controlled the orientation of quadrangle patches and confirmed a significantly larger reduction in the strength of the illusion when the edge orientations of quadrangle patches were vertical and horizontal (square) than when they were oblique (diamond). To explore the relationship between orientation processing and scintillating grid illusion, we controlled, in experiment 3, the global orientation of the display; the strength of the illusion with diamonds was significantly weaker when it was rotated by 45 degrees than when it was not rotated. These results indicate that it is not only the difference of edge orientation of luminance patches, but also the orientation with respect to the grid that determines the strength of the illusion.

Artificial neural network-based glaucoma diagnosis using retinal nerve fiber layer analysis

PURPOSE

To develop, train, and test an artificial neural network (ANN) for differentiating among normal subjects, primary open angle glaucoma (POAG) suspects, and persons with POAG in Asian-Indian eyes using inputs from clinical parameters, optical coherence tomography (OCT), visual fields, and GDx nerve fiber analyzer.

METHODS

One hundred eyes were classified using optic disc examination and perimetry into normal (n=35), POAG suspects (n=30), and POAG (n=35). EasyNN-plus simulator was used to develop an ANN model with inputs including age, sex, myopia, intraocular pressure (IOP), optic nerve head, and retinal nerve fiber layer (RNFL) parameters on OCT, Octopus 30-2 full threshold visual field, and GDx parameters.

RESULTS

With two outputs (POAG or normal), specificity was 80% and sensitivity was 93.3%. Ninety percent of POAG suspects were labeled as abnormal in this analysis. ANN assigned the highest importance to Smax/Imax RNFL on OCT followed by cup-area (OCT) and other RNFL parameters (OCT) for two outputs. With three outputs (normal, POAG, and POAG suspect), ANN gave an overall classification rate of 65%, specificity of 60%, and sensitivity of 71.4% with a target error rate of the training set at 1%. The parameters for three outputs, in decreasing order of relative importance, were Savg, vertical cup-disc ratio, cup-volume, and cup-area on OCT.

CONCLUSIONS

An ANN taking varied diagnostic imaging inputs was able to separate POAG eyes from normal subjects and POAG suspects. The network had reasonable sensitivity with three outputs; however, it had a tendency to mislabel POAG suspects as POAG.

[The Hermann grid illusion: the classic textbook interpretation is obsolete]

The Hermann grid is an optical illusion in which the crossings of white grid lines appear darker than the grid lines outside the crossings. The illusion disappears when one fixates the crossings. The discoverer, Ludimar Hermann (1838-1914), interpreted the illusion as evidence for lateral connections in the retina. In most textbooks on sensory physiology and ophthalmology, the Hermann grid illusion serves to illustrate "lateral inhibition." This paper summarises new findings that show that the classic explanation is incomplete. In 2004, a seemingly subtle modification, a small undulation of the grid lines, was shown to demolish the illusion. In 2007, a more convincing explanation appeared: An artificial neural network was trained for "lightness constancy"- the ability of our visual system to interpret luminance in the interest of object recognition, independent of illumination. After having learned lightness constancy, the network was subjected to a number of lightness illusions, among them the Hermann grid illusion. An analysis of the coupling constants of this neural network promises to further our understanding of the Hermann grid illusion.