Detecting Glaucoma from Fundus Photographs Using Deep Learning without Convolutions: Transformer for Improved Generalization

• Published in: Ophthalmology science (Online) Vol. 3; no. 1; p. 100233
• Main Authors: Fan, Rui, Alipour, Kamran, Bowd, Christopher, Christopher, Mark, Brye, Nicole, Proudfoot, James A, Goldbaum, Michael H, Belghith, Akram, Girkin, Christopher A, Fazio, Massimo A, Liebmann, Jeffrey M, Weinreb, Robert N, Pazzani, Michael, Kriegman, David, Zangwill, Linda M
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier 01.03.2023
• Subjects: Artificial Intelligence and Big Data; Ophthalmology; Fundus photographs; CI, confidence interval; OHTS, Ocular Hypertension Treatment Study; AUROC, areas under the receiver operating characteristic curve; DL, deep learning; CNN, convolutional neural network; Deep learning; AI, artificial intelligence; VF, visual field; LAG, Large-Scale Attention-Based Glaucoma; DeiT, Data-efficient image Transformer; SoTA, state-of-the-art; POAG, primary open-angle glaucoma; Vision Transformers; ViT, Vision Transformer; Glaucoma detection

AbstractPurposeTo compare the diagnostic accuracy and explainability of a new Vision Transformer deep learning technique, Data-efficient image Transformer (DeiT), and Resnet-50, trained on fundus photographs from the Ocular Hypertension Treatment Study (OHTS) to detect primary open-angle glaucoma (P...