A deep learning model for the detection of both advanced and early glaucoma using fundus photography

• Published in: PloS one Vol. 13; no. 11; p. e0207982
• Main Authors: Ahn, Jin Mo, Kim, Sangsoo, Ahn, Kwang-Sung, Cho, Sung-Hoon, Lee, Kwan Bok, Kim, Ungsoo Samuel
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 27.11.2018; Public Library of Science (PLoS)
• Subjects: Analysis; Artificial neural networks; Biology and Life Sciences; Case studies; Classification; Computer and Information Sciences; Cross-Sectional Studies; Data transfer (computers); Datasets; Deep Learning; Diabetic retinopathy; Diagnosis; Diagnostic Techniques, Ophthalmological; Disease Progression; Fundus Oculi; Glaucoma; Glaucoma - diagnostic imaging; Humans; Image Interpretation, Computer-Assisted - methods; Learning algorithms; Logistic Models; Machine learning; Macular degeneration; Medicine and Health Sciences; Model accuracy; Neural networks; Photography; Physical Sciences; Research and Analysis Methods; ROC Curve; Training; Japan
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0207982

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.