Validation of a Deep Learning Model to Screen for Glaucoma Using Images from Different Fundus Cameras and Data Augmentation

Effects of Study Population, Labeling and Training on Glaucoma Detection Using Deep Learning Algorithms

“…Indeed, our research teams from UCSD and UTokyo recently reported the benefits of transfer learning using Imagenet in glaucoma detection from fundus photographs as well as diagnosing early stage glaucoma from optical coherence tomography (OCT) images. 5,6,25,26 In both cases, diagnostic performance of the deep learning model was significantly improved by pretraining using transfer learning. 5,6 Updating the trained model with new images for the same specialized task can also improve performance.…”

“…To address this issue, various groups around the world, using different photographic techniques and deep learning strategies, 4 have proposed methods for automated evaluation of fundus photographs to detect glaucoma. [5][6][7][8][9][10][11][12][13] We recently have shown that these approaches can be effective despite differences in fundus camera resolution capability or sensor type. 5 Most important, deep learning-based methods have been shown to achieve high accuracy on unseen datasets, thanks to their ability to use complex visual features in fundus photographs for the assessment.…”

“…[5][6][7][8][9][10][11][12][13] We recently have shown that these approaches can be effective despite differences in fundus camera resolution capability or sensor type. 5 Most important, deep learning-based methods have been shown to achieve high accuracy on unseen datasets, thanks to their ability to use complex visual features in fundus photographs for the assessment. [5][6][7][8][9][10][11][12][13] Currently, however, little is known about how generalizable these deep learning algorithms for glaucoma detection are to new and independent patient populations and how well different models perform on the same datasets.…”

“…5 Most important, deep learning-based methods have been shown to achieve high accuracy on unseen datasets, thanks to their ability to use complex visual features in fundus photographs for the assessment. [5][6][7][8][9][10][11][12][13] Currently, however, little is known about how generalizable these deep learning algorithms for glaucoma detection are to new and independent patient populations and how well different models perform on the same datasets. Often, models are developed and trained on data collected from a specific geographic region or from a homogeneous population and their performance on different populations (e.g., patients from racial groups that are not represented in the training data) is not well-characterized.…”

Visualizing the dynamic change of Ocular Response Analyzer waveform using Variational Autoencoder in association with the peripapillary retinal arteries angle

“…In machine learning, including deep learning, discriminative and generative models are used 17 . Several studies suggested the usefulness of a discriminative deep learning approach in Ophthalmology, for example for diagnosing glaucoma from a fundus photography [18][19][20][21][22] and from optical coherence tomography (OCT) 23 . Variational Autoencoders (VAEs) are a type of deep learning approach that allows powerful generative models of data 24,25 .…”