Evaluation of deep convolutional neural networks for glaucoma detection

The Japan Ocular Imaging Registry Research Group

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

Purpose: To investigate the performance of deep convolutional neural networks (DCNNs) for glaucoma discrimination using color fundus images Study design: A retrospective study Patients and methods: To investigate the discriminative ability of 3 DCNNs, we used a total of 3312 images consisting of 369 images from glaucoma-confirmed eyes, 256 images from glaucoma-suspected eyes diagnosed by a glaucoma expert, and 2687 images judged to be nonglaucomatous eyes by a glaucoma expert. We also investigated the effects of image size on the discriminative ability and heatmap analysis to determine which parts of the image contribute to the discrimination. Additionally, we used 465 poor-quality images to investigate the effect of poor image quality on the discriminative ability. Results: Three DCNNs showed areas under the curve (AUCs) of 0.9 or more. The AUC of the DCNN using glaucoma-confirmed eyes against nonglaucomatous eyes was higher than that using glaucoma-suspected eyes against nonglaucomatous eyes by approximately 0.1. The image size did not affect the discriminative ability. Heatmap analysis showed that the optic disc area was the most important area for the discrimination of glaucoma. The image quality affected the discriminative ability, and the inclusion of poor-quality images in the analysis reduced the AUC by 0.1 to 0.2. Conclusions: DCNNs may be a useful tool for detecting glaucoma or glaucoma-suspected eyes by use of fundus color images. Proper preprocessing and collection of qualified images are essential to improving the discriminative ability.

Original languageEnglish
Pages (from-to)276-283
Number of pages8
JournalJapanese Journal of Ophthalmology
Volume63
Issue number3
DOIs
Publication statusPublished - May 16 2019

Fingerprint

Glaucoma
Area Under Curve
Color
Optic Disk
Retrospective Studies

All Science Journal Classification (ASJC) codes

  • Ophthalmology

Cite this

Evaluation of deep convolutional neural networks for glaucoma detection. / The Japan Ocular Imaging Registry Research Group.

In: Japanese Journal of Ophthalmology, Vol. 63, No. 3, 16.05.2019, p. 276-283.

Research output: Contribution to journalArticle

The Japan Ocular Imaging Registry Research Group. / Evaluation of deep convolutional neural networks for glaucoma detection. In: Japanese Journal of Ophthalmology. 2019 ; Vol. 63, No. 3. pp. 276-283.
@article{b45abab3288f46158d3cc02927c3227b,
title = "Evaluation of deep convolutional neural networks for glaucoma detection",
abstract = "Purpose: To investigate the performance of deep convolutional neural networks (DCNNs) for glaucoma discrimination using color fundus images Study design: A retrospective study Patients and methods: To investigate the discriminative ability of 3 DCNNs, we used a total of 3312 images consisting of 369 images from glaucoma-confirmed eyes, 256 images from glaucoma-suspected eyes diagnosed by a glaucoma expert, and 2687 images judged to be nonglaucomatous eyes by a glaucoma expert. We also investigated the effects of image size on the discriminative ability and heatmap analysis to determine which parts of the image contribute to the discrimination. Additionally, we used 465 poor-quality images to investigate the effect of poor image quality on the discriminative ability. Results: Three DCNNs showed areas under the curve (AUCs) of 0.9 or more. The AUC of the DCNN using glaucoma-confirmed eyes against nonglaucomatous eyes was higher than that using glaucoma-suspected eyes against nonglaucomatous eyes by approximately 0.1. The image size did not affect the discriminative ability. Heatmap analysis showed that the optic disc area was the most important area for the discrimination of glaucoma. The image quality affected the discriminative ability, and the inclusion of poor-quality images in the analysis reduced the AUC by 0.1 to 0.2. Conclusions: DCNNs may be a useful tool for detecting glaucoma or glaucoma-suspected eyes by use of fundus color images. Proper preprocessing and collection of qualified images are essential to improving the discriminative ability.",
author = "{The Japan Ocular Imaging Registry Research Group} and Sang Phan and Shin’ichi Satoh and Yoshioki Yoda and Kenji Kashiwagi and Tetsuro Oshika and Takashi Hasegawa and Kenji Kashiwagi and Masahiro Miyake and Taiji Sakamoto and Takeshi Yoshitomi and Masaru Inatani and Tetsuya Yamamoto and Kazuhisa Sugiyama and Makoto Nakamura and Akitaka Tsujikawa and Chie Sotozono and Kohei Sonoda and Hiroko Terasaki and Yuichiro Ogura and Takeo Fukuchi and Fumio Shiraga and Kohji Nishida and Toru Nakazawa and Makoto Aihara and Hidetoshi Yamashita and Iijima Hiyoyuki",
year = "2019",
month = "5",
day = "16",
doi = "10.1007/s10384-019-00659-6",
language = "English",
volume = "63",
pages = "276--283",
journal = "Japanese Journal of Ophthalmology",
issn = "0021-5155",
publisher = "Springer Japan",
number = "3",

}

TY - JOUR

T1 - Evaluation of deep convolutional neural networks for glaucoma detection

AU - The Japan Ocular Imaging Registry Research Group

AU - Phan, Sang

AU - Satoh, Shin’ichi

AU - Yoda, Yoshioki

AU - Kashiwagi, Kenji

AU - Oshika, Tetsuro

AU - Hasegawa, Takashi

AU - Kashiwagi, Kenji

AU - Miyake, Masahiro

AU - Sakamoto, Taiji

AU - Yoshitomi, Takeshi

AU - Inatani, Masaru

AU - Yamamoto, Tetsuya

AU - Sugiyama, Kazuhisa

AU - Nakamura, Makoto

AU - Tsujikawa, Akitaka

AU - Sotozono, Chie

AU - Sonoda, Kohei

AU - Terasaki, Hiroko

AU - Ogura, Yuichiro

AU - Fukuchi, Takeo

AU - Shiraga, Fumio

AU - Nishida, Kohji

AU - Nakazawa, Toru

AU - Aihara, Makoto

AU - Yamashita, Hidetoshi

AU - Hiyoyuki, Iijima

PY - 2019/5/16

Y1 - 2019/5/16

N2 - Purpose: To investigate the performance of deep convolutional neural networks (DCNNs) for glaucoma discrimination using color fundus images Study design: A retrospective study Patients and methods: To investigate the discriminative ability of 3 DCNNs, we used a total of 3312 images consisting of 369 images from glaucoma-confirmed eyes, 256 images from glaucoma-suspected eyes diagnosed by a glaucoma expert, and 2687 images judged to be nonglaucomatous eyes by a glaucoma expert. We also investigated the effects of image size on the discriminative ability and heatmap analysis to determine which parts of the image contribute to the discrimination. Additionally, we used 465 poor-quality images to investigate the effect of poor image quality on the discriminative ability. Results: Three DCNNs showed areas under the curve (AUCs) of 0.9 or more. The AUC of the DCNN using glaucoma-confirmed eyes against nonglaucomatous eyes was higher than that using glaucoma-suspected eyes against nonglaucomatous eyes by approximately 0.1. The image size did not affect the discriminative ability. Heatmap analysis showed that the optic disc area was the most important area for the discrimination of glaucoma. The image quality affected the discriminative ability, and the inclusion of poor-quality images in the analysis reduced the AUC by 0.1 to 0.2. Conclusions: DCNNs may be a useful tool for detecting glaucoma or glaucoma-suspected eyes by use of fundus color images. Proper preprocessing and collection of qualified images are essential to improving the discriminative ability.

AB - Purpose: To investigate the performance of deep convolutional neural networks (DCNNs) for glaucoma discrimination using color fundus images Study design: A retrospective study Patients and methods: To investigate the discriminative ability of 3 DCNNs, we used a total of 3312 images consisting of 369 images from glaucoma-confirmed eyes, 256 images from glaucoma-suspected eyes diagnosed by a glaucoma expert, and 2687 images judged to be nonglaucomatous eyes by a glaucoma expert. We also investigated the effects of image size on the discriminative ability and heatmap analysis to determine which parts of the image contribute to the discrimination. Additionally, we used 465 poor-quality images to investigate the effect of poor image quality on the discriminative ability. Results: Three DCNNs showed areas under the curve (AUCs) of 0.9 or more. The AUC of the DCNN using glaucoma-confirmed eyes against nonglaucomatous eyes was higher than that using glaucoma-suspected eyes against nonglaucomatous eyes by approximately 0.1. The image size did not affect the discriminative ability. Heatmap analysis showed that the optic disc area was the most important area for the discrimination of glaucoma. The image quality affected the discriminative ability, and the inclusion of poor-quality images in the analysis reduced the AUC by 0.1 to 0.2. Conclusions: DCNNs may be a useful tool for detecting glaucoma or glaucoma-suspected eyes by use of fundus color images. Proper preprocessing and collection of qualified images are essential to improving the discriminative ability.

UR - http://www.scopus.com/inward/record.url?scp=85062148105&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85062148105&partnerID=8YFLogxK

U2 - 10.1007/s10384-019-00659-6

DO - 10.1007/s10384-019-00659-6

M3 - Article

C2 - 30798379

AN - SCOPUS:85062148105

VL - 63

SP - 276

EP - 283

JO - Japanese Journal of Ophthalmology

JF - Japanese Journal of Ophthalmology

SN - 0021-5155

IS - 3

ER -