2020
DOI: 10.1371/journal.pone.0233079
|View full text |Cite
|
Sign up to set email alerts
|

Deep learning assisted detection of glaucomatous optic neuropathy and potential designs for a generalizable model

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. Methods 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 wa… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
8
0

Year Published

2021
2021
2022
2022

Publication Types

Select...
7
2

Relationship

1
8

Authors

Journals

citations
Cited by 25 publications
(8 citation statements)
references
References 23 publications
0
8
0
Order By: Relevance
“…We used an ensemble approach by incorporating a support vector machine to identify discs with enlarged cup-to-disc ratio when the confidence score of the CNN classifier was less than 0.85. However, this approach is still limited by the overlapped distribution of cup-to-disc ratio among the healthy and glaucoma subjects [ 32 ]. Increasing the number and variability of the training images is a potential way to increase the generalizability of a DL model, but its effect reached plateau when using 60,000 or more training images in the DR model, and is suboptimal in glaucoma models [ 14 , 33 ].…”
Section: Discussionmentioning
confidence: 99%
“…We used an ensemble approach by incorporating a support vector machine to identify discs with enlarged cup-to-disc ratio when the confidence score of the CNN classifier was less than 0.85. However, this approach is still limited by the overlapped distribution of cup-to-disc ratio among the healthy and glaucoma subjects [ 32 ]. Increasing the number and variability of the training images is a potential way to increase the generalizability of a DL model, but its effect reached plateau when using 60,000 or more training images in the DR model, and is suboptimal in glaucoma models [ 14 , 33 ].…”
Section: Discussionmentioning
confidence: 99%
“…Ko et al (32) evaluated and detected the performance of a CNN framework for vertical CDR (VCDR) classification using 944 fundus photographs, including 465 nonglaucomatous optic neuropathy (NGON) eyes and 479 glaucomatous optic neuropathy (GON) eyes. Based on stratified sampling, the images were divided into a training set consisting of 763 images and a test set consisting of 181 images.…”
Section: Segmentation Of the Optic Cup And The Optic Discmentioning
confidence: 99%
“…Model [20]. The primary challenge of the development is applying neural networks in the feature extraction process.…”
Section: 21mentioning
confidence: 99%