Ryo Asaoka1, Hiroshi Murata2, Kazunori Hirasawa3, Yuri Fujino2, Masato Matsuura4, Atsuya Miki5, Takashi Kanamoto6, Yoko Ikeda7, Kazuhiko Mori8, Aiko Iwase8, Nobuyuki Shoji9, Kenji Inoue10, Junkichi Yamagami11, Makoto Araie12. 1. Department of Ophthalmology, The University of Tokyo, Tokyo, Japan. Electronic address: rasaoka-tky@umin.ac.jp. 2. Department of Ophthalmology, The University of Tokyo, Tokyo, Japan. 3. Moorfields Eye Hospital National Health Service Foundation Trust and University College London, Institute of Ophthalmology, London, United Kingdom; Department of Ophthalmology, School of Medicine, Kitasato University, Kanagawa, Japan. 4. Department of Ophthalmology, The University of Tokyo, Tokyo, Japan; Moorfields Eye Hospital National Health Service Foundation Trust and University College London, Institute of Ophthalmology, London, United Kingdom. 5. Department of Ophthalmology, Osaka University Graduate School of Medicine, Osaka, Japan. 6. Department of Ophthalmology, Hiroshima Memorial Hospital, Hiroshima, Japan. 7. Department of Ophthalmology, Kyoto Prefectural University of Medicine, Kyoto, Japan; Oike Ikeda Eye Clinic, Kyoto, Japan. 8. Tajimi Iwase Eye Clinic, Tajimi, Japan. 9. Department of Ophthalmology, School of Medicine, Kitasato University, Kanagawa, Japan. 10. Inouye Eye Hospital, Tokyo, Japan. 11. JR Tokyo General Hospital, Tokyo, Japan. 12. Kanto Central Hospital of the Mutual Aid Association of Public School Teachers, Tokyo, Japan.
Abstract
PURPOSE: We sought to construct and evaluate a deep learning (DL) model to diagnose early glaucoma from spectral-domain optical coherence tomography (OCT) images. DESIGN: Artificial intelligence diagnostic tool development, evaluation, and comparison. METHODS: This multi-institution study included pretraining data of 4316 OCT images (RS3000) from 1371 eyes with open angle glaucoma (OAG) regardless of the stage of glaucoma and 193 normal eyes. Training data included OCT-1000/2000 images from 94 eyes of 94 patients with early OAG (mean deviation > -5.0 dB) and 84 eyes of 84 normal subjects. Testing data included OCT-1000/2000 from 114 eyes of 114 patients with early OAG (mean deviation > -5.0 dB) and 82 eyes of 82 normal subjects. A DL (convolutional neural network) classifier was trained using a pretraining dataset, followed by a second round of training using an independent training dataset. The DL model input features were the 8 × 8 grid macular retinal nerve fiber layer thickness and ganglion cell complex layer thickness from spectral-domain OCT. Diagnostic accuracy was investigated in the testing dataset. For comparison, diagnostic accuracy was also evaluated using the random forests and support vector machine models. The primary outcome measure was the area under the receiver operating characteristic curve (AROC). RESULTS: The AROC with the DL model was 93.7%. The AROC significantly decreased to between 76.6% and 78.8% without the pretraining process. Significantly smaller AROCs were obtained with random forests and support vector machine models (82.0% and 67.4%, respectively). CONCLUSION: A DL model for glaucoma using spectral-domain OCT offers a substantive increase in diagnostic performance.
PURPOSE: We sought to construct and evaluate a deep learning (DL) model to diagnose early glaucoma from spectral-domain optical coherence tomography (OCT) images. DESIGN: Artificial intelligence diagnostic tool development, evaluation, and comparison. METHODS: This multi-institution study included pretraining data of 4316 OCT images (RS3000) from 1371 eyes with open angle glaucoma (OAG) regardless of the stage of glaucoma and 193 normal eyes. Training data included OCT-1000/2000 images from 94 eyes of 94 patients with early OAG (mean deviation > -5.0 dB) and 84 eyes of 84 normal subjects. Testing data included OCT-1000/2000 from 114 eyes of 114 patients with early OAG (mean deviation > -5.0 dB) and 82 eyes of 82 normal subjects. A DL (convolutional neural network) classifier was trained using a pretraining dataset, followed by a second round of training using an independent training dataset. The DL model input features were the 8 × 8 grid macular retinal nerve fiber layer thickness and ganglion cell complex layer thickness from spectral-domain OCT. Diagnostic accuracy was investigated in the testing dataset. For comparison, diagnostic accuracy was also evaluated using the random forests and support vector machine models. The primary outcome measure was the area under the receiver operating characteristic curve (AROC). RESULTS: The AROC with the DL model was 93.7%. The AROC significantly decreased to between 76.6% and 78.8% without the pretraining process. Significantly smaller AROCs were obtained with random forests and support vector machine models (82.0% and 67.4%, respectively). CONCLUSION: A DL model for glaucoma using spectral-domain OCT offers a substantive increase in diagnostic performance.
Authors: Eduardo M Normando; Tim E Yap; John Maddison; Serge Miodragovic; Paolo Bonetti; Melanie Almonte; Nada G Mohammad; Sally Ameen; Laura Crawley; Faisal Ahmed; Philip A Bloom; Maria Francesca Cordeiro Journal: Expert Rev Mol Diagn Date: 2020-05-03 Impact factor: 5.225
Authors: Atalie C Thompson; Alessandro A Jammal; Samuel I Berchuck; Eduardo B Mariottoni; Felipe A Medeiros Journal: JAMA Ophthalmol Date: 2020-04-01 Impact factor: 7.389
Authors: Mengyu Wang; Lucy Q Shen; Louis R Pasquale; Michael V Boland; Sarah R Wellik; Carlos Gustavo De Moraes; Jonathan S Myers; Thao D Nguyen; Robert Ritch; Pradeep Ramulu; Hui Wang; Jorryt Tichelaar; Dian Li; Peter J Bex; Tobias Elze Journal: Ophthalmology Date: 2019-12-12 Impact factor: 12.079
Authors: Andrew C Lin; Cecilia S Lee; Marian Blazes; Aaron Y Lee; Michael B Gorin Journal: Transl Vis Sci Technol Date: 2021-05-03 Impact factor: 3.283
Authors: An Ran Ran; Clement C Tham; Poemen P Chan; Ching-Yu Cheng; Yih-Chung Tham; Tyler Hyungtaek Rim; Carol Y Cheung Journal: Eye (Lond) Date: 2020-10-07 Impact factor: 3.775
Authors: Gala Beykin; Anthony M Norcia; Vivek J Srinivasan; Alfredo Dubra; Jeffrey L Goldberg Journal: Prog Retin Eye Res Date: 2020-07-10 Impact factor: 21.198