Deep learning assisted detection of glaucomatous optic neuropathy and potential designs for a generalizable model
Open Access
- 14 May 2020
- journal article
- research article
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 15 (5) , e0233079
Abstract
To evaluate ways to improve the generalizability of a deep learning algorithm for identifying glaucomatous optic neuropathy (GON) using a limited number of fundus photographs, as well as the key features being used for classification. A total of 944 fundus images from Taipei Veterans General Hospital (TVGH) were retrospectively collected. Clinical and demographic characteristics, including structural and functional measurements of the images with GON, were recorded. Transfer learning based on VGGNet was used to construct a convolutional neural network (CNN) to identify GON. To avoid missing cases with advanced GON, an ensemble model was adopted in which a support vector machine classifier would make final classification based on cup-to-disc ratio if the CNN classifier had low-confidence score. The CNN classifier was first established using TVGH dataset, and then fine-tuned by combining the training images of TVGH and Drishti-GS datasets. Class activation map (CAM) was used to identify key features used for CNN classification. Performance of each classifier was determined through area under receiver operating characteristic curve (AUC) and compared with the ensemble model by diagnostic accuracy. In 187 TVGH test images, the accuracy, sensitivity, and specificity of the CNN classifier were 95.0%, 95.7%, and 94.2%, respectively, and the AUC was 0.992 compared to the 92.8% accuracy rate of the ensemble model. For the Drishti-GS test images, the accuracy of the CNN, the fine-tuned CNN and ensemble model was 33.3%, 80.3%, and 80.3%, respectively. The CNN classifier did not misclassify images with moderate to severe diseases. Class-discriminative regions revealed by CAM co-localized with known characteristics of GON. The ensemble model or a fine-tuned CNN classifier may be potential designs to build a generalizable deep learning model for glaucoma detection when large image databases are not available.Keywords
Funding Information
- Ministry of Science and Technology, Taiwan (MOST 106-2314-B-075-037-& MOST 105-2314-B-075-047)
- Taipei Veterans General Hospital (V107C-114)
- Ministry of Science and Technology, Taiwan (MOST 108-3011-F-075-001 and MOST 107-2634-F-075-001)
This publication has 24 references indexed in Scilit:
- The Effectiveness of Teleglaucoma versus In-Patient Examination for Glaucoma Screening: A Systematic Review and Meta-AnalysisPLOS ONE, 2014
- Glaucomatous Optic Neuropathy Evaluation ProjectJAMA Ophthalmology, 2014
- Automatic extraction of retinal features from colour retinal images for glaucoma diagnosis: A reviewComputerized Medical Imaging and Graphics, 2013
- An Evidence-Based Review of Prognostic Factors for Glaucomatous Visual Field ProgressionOphthalmology, 2013
- Optic Disk and Nerve Fiber Layer Imaging to Detect GlaucomaAmerican Journal of Ophthalmology, 2007
- Blindness in patients with treated open-angle glaucomaOphthalmology, 2003
- Vertical cup/disc ratio in relation to optic disc size: its value in the assessment of the glaucoma suspectBritish Journal of Ophthalmology, 1998
- Prevalence of Open-angle Glaucoma in AustraliaOphthalmology, 1996
- Glaucoma Care and Conformance with Preferred Practice PatternsOphthalmology, 1996
- The Prevalence of Primary Open-angle Glaucoma in a Population-based Study in The NetherlandsOphthalmology, 1994