To build a deep learning model to diagnose glaucoma using fundus photography. Cross sectional case study Subjects, Participants and Controls: A total of 1,542 photos (786 normal controls, 467 advanced glaucoma and 289 early glaucoma patients) were obtained by fundus photography. The whole dataset of 1,542 images were split into 754 training, 324 validation and 464 test datasets. These datasets were used to construct simple logistic classification and convolutional neural network using Tensorflow. The same datasets were used to fine tune pre-trained GoogleNet Inception v3 model. The simple logistic classification model showed a training accuracy of 82.9%, validation accuracy of 79.9% and test accuracy of 77.2%. Convolutional neural network achieved accuracy and area under the receiver operating characteristic curve (AUROC) of 92.2% and 0.98 on the training data, 88.6% and 0.95 on the validation data, and 87.9% and 0.94 on the test data. Transfer-learned GoogleNet Inception v3 model achieved accuracy and AUROC of 99.7% and 0.99 on training data, 87.7% and 0.95 on validation data, and 84.5% and 0.93 on test data. Both advanced and early glaucoma could be correctly detected via machine learning, using only fundus photographs. Our new model that is trained using convolutional neural network is more efficient for the diagnosis of early glaucoma than previously published models.

Retinal ganglion cells (RGCs) are the bridging neurons that connect the retinal input to the visual processing centres within the central nervous system. There is a remarkable diversity of RGCs and the various subtypes have unique morphological features, distinct functions, and characteristic pathways linking the inner retina to the relevant brain areas. A number of psychophysical and electrophysiological tests have been refined to investigate this large and varied population of RGCs. Technological advances, such as high-resolution optical coherence tomography imaging, have provided additional tools to define the pattern of RGC involvement and the chronological sequence of events in both inherited and acquired optic neuropathies. The mechanistic insights gained from these studies, in particular the selective vulnerability and relative resilience of particular RGC subtypes, are of fundamental importance as they are directly relevant to the development of targeted therapies for these invariably progressive blinding diseases. This review provides a comprehensive description of the various types of RGCs, the developments in proposed methods of classification, and the current gaps in our knowledge of how these RGCs are differentially affected depending on the underlying aetiology. The synthesis of the current body of knowledge on the diversity of RGCs and the pathways that are potentially amenable to therapeutic modulation will hopefully lead to much needed effective treatments for patients with optic neuropathies.

This study is to evaluate the accuracy of machine learning for differentiation between optic neuropathies, pseudopapilledema (PPE) and normals. Two hundred and ninety-five images of optic neuropathies, 295 images of PPE, and 779 control images were used. Pseudopapilledema was defined as follows: cases with elevated optic nerve head and blurred disc margin, with normal visual acuity (> 0.8 Snellen visual acuity), visual field, color vision, and pupillary reflex. The optic neuropathy group included cases of ischemic optic neuropathy (177), optic neuritis (48), diabetic optic neuropathy (17), papilledema (22), and retinal disorders (31). We compared four machine learning classifiers (our model, GoogleNet Inception v3, 19-layer Very Deep Convolution Network from Visual Geometry group (VGG), and 50-layer Deep Residual Learning (ResNet)). Accuracy and area under receiver operating characteristic curve (AUROC) were analyzed. The accuracy of machine learning classifiers ranged from 95.89 to 98.63% (our model: 95.89%, Inception V3: 96.45%, ResNet: 98.63%, and VGG: 96.80%). A high AUROC score was noted in both ResNet and VGG (0.999). Machine learning techniques can be combined with fundus photography as an effective approach to distinguish between PPE and elevated optic disc associated with optic neuropathies.

Jung JJ, et al. Korean J Ophthalmol. 2011 Feb;25(1):33-36. https://doi.org/10.3341/kjo.2011.25.1.33

Purpose: To compare a biometer using swept-source optical coherence tomography (SS-OCT) with a partial coherence interferometry (PCI)-based biometer in measurements of two ocular biometry parameters, i.e., the axial length and anterior cornea curvature. Methods: We compared the two biometers SS-OCT (ANTERION, Heidelberg Engineering Inc., Heidelberg, Germany) and PCI (IOL Master, Carl Zeiss Meditec, Jena, Germany) in terms of the axial length (AL) and corneal curvature (K) measurements of 175 eyes. Paired t-tests were used to compare the two biometers. Agreement between the biometers was evaluated using the Bland–Altman method. Results: The mean age was 36.0 ± 25.6 years (range: 5 to 85 years). The mean axial length was 24.42 ± 0.13 mm for SS-OCT and 24.45 ± 0.14 mm for PCI. The mean corneal curvature was significantly different between the two biometry in flat K (K1) but not in steep K (K2). The limit of agreement was -0.15 to 0.21 in the axial length, -1.18 to 0.83 in K1, and -1.06 to 0.95 in K2. All above ocular biometric measurements between SS-OCT and PCI correlated significantly (Pearson's correlation, p<0.001). Conclusions: The axial length measured using SS-OCT is useful in clinical practice. It shows a good correlation and agreement with that measured using PCI. However, the axial length and corneal curvature measured using SS-OCT cannot be used interchangeably with that measured using PCI in clinical practice.