Automatic Detection of Glaucomatous Visual Field Progression With Neural Networks

Neural Network Analysis of Different Segmentation Strategies of Nerve Fiber Layer Assessment for Glaucoma Diagnosis

“…Artificial neural networks have been successfully applied to interpret and classify visual fields, 13,14 detect visual field progression, 15 assess structural data from the optic nerve head, 16,17 and determine which machine-learning classifier Areas under the ROC curves (AUC), confidence intervals (CI), sensitivity, specificity, and likelihood ratios of artificial neural network and retinal nerve fiber layer parameters of OCT Spectralis to discriminate between normal and glaucoma subjects depending on the segmentation strategy.…”

Detection of visual field progression in glaucoma with standard achromatic perimetry: A review and practical implications

“…This model could be used to classify longitudinal visual field data, which may prove valuable in monitoring visual field loss. Another early study evaluated 233 visual field series of patients with glaucoma experienced in automated perimetry to identify visual field progression with neural networks [88]. Agreement of three experienced observers was used as the "gold standard".…”

Glaucoma detection and evaluation through pattern recognition in standard automated perimetry data

“…In the study by Anton et al [16], the logistic discriminant analysis correctly identified the glaucomatous or non-glaucomatous defects, with a sensitivity of 65% to 85% and a specificity of 60% These results indicate the importance of ancillary data in glaucoma diagnosis. When used for automatic detection of glaucomatous visual field progression, the MLP neural network showed 73% sensitivity with 88% specificity, and a reasonably good agreement with the opinions of human experts [19]. Lietman et al [20] compared the glaucoma detection performance of a MLP network with that of ten other methods, including the global indices (e.g., mean deviation and corrected pattern standard deviation), methods to identify clusters of abnormal points, and cross-meridional algorithms.…”