Retinal image quality classification using fine-tuned CNN

Jing Sun; Cheng Wan; Jun Cheng; Fengli Yu; Jiang Liu

doi:10.1007/978-3-319-67561-9_14

Retinal image quality classification using fine-tuned CNN

Jing Sun
, Cheng Wan^*
, Jun Cheng
, Fengli Yu
, Jiang Liu

^*Corresponding author for this work

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

14 Citations (Scopus)

Abstract

Retinal image quality classification makes a great difference in automated diabetic retinopathy screening systems. With the increase of application of portable fundus cameras, we can get a large number of retinal images, but there are quite a number of images in poor quality because of uneven illumination, occlusion and patients movements. Using the dataset with poor quality training networks for DR screening system will lead to the decrease of accuracy. In this paper, we first explore four CNN architectures (AlexNet, GoogLeNet, VGG-16, and ResNet-50) from ImageNet image classification task to our Retinal fundus images quality classification, then we pick top two networks out and jointly fine-tune the two networks. The total loss of the network we proposed is equal to the sum of the losses of all channels. We demonstrate the super performance of our proposed algorithm on a large retinal fundus image dataset and achieve an optimal accuracy of 97.12%, outperforming the current methods in this area.

Original language	English
Title of host publication	Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings
Editors	M. Jorge Cardoso, Tal Arbel
Publisher	Springer Verlag
Pages	126-133
Number of pages	8
ISBN (Print)	9783319675602
DOIs	https://doi.org/10.1007/978-3-319-67561-9_14
Publication status	Published - 2017
Externally published	Yes
Event	International Workshop on Fetal and Infant Image Analysis, FIFI 2017 and 4th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2017 held in Conjunction with 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017 - Quebec City, Canada Duration: 14 Sept 2017 → 14 Sept 2017

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	10554 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	International Workshop on Fetal and Infant Image Analysis, FIFI 2017 and 4th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2017 held in Conjunction with 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017
Country/Territory	Canada
City	Quebec City
Period	14/09/17 → 14/09/17

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 3 Good Health and Well-being

Free Keywords

Convolutional neural networks (CNN)
Fine-tuning
No-reference image quality assessment (NR-IQA)
Retinal image

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-319-67561-9_14

Cite this

Sun, J., Wan, C., Cheng, J., Yu, F., & Liu, J. (2017). Retinal image quality classification using fine-tuned CNN. In M. J. Cardoso, & T. Arbel (Eds.), Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings (pp. 126-133). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10554 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-67561-9_14

Sun, Jing ; Wan, Cheng ; Cheng, Jun et al. / Retinal image quality classification using fine-tuned CNN. Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings. editor / M. Jorge Cardoso ; Tal Arbel. Springer Verlag, 2017. pp. 126-133 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{33e547eafab440eca79398da7e0891ec,

title = "Retinal image quality classification using fine-tuned CNN",

abstract = "Retinal image quality classification makes a great difference in automated diabetic retinopathy screening systems. With the increase of application of portable fundus cameras, we can get a large number of retinal images, but there are quite a number of images in poor quality because of uneven illumination, occlusion and patients movements. Using the dataset with poor quality training networks for DR screening system will lead to the decrease of accuracy. In this paper, we first explore four CNN architectures (AlexNet, GoogLeNet, VGG-16, and ResNet-50) from ImageNet image classification task to our Retinal fundus images quality classification, then we pick top two networks out and jointly fine-tune the two networks. The total loss of the network we proposed is equal to the sum of the losses of all channels. We demonstrate the super performance of our proposed algorithm on a large retinal fundus image dataset and achieve an optimal accuracy of 97.12\%, outperforming the current methods in this area.",

keywords = "Convolutional neural networks (CNN), Fine-tuning, No-reference image quality assessment (NR-IQA), Retinal image",

author = "Jing Sun and Cheng Wan and Jun Cheng and Fengli Yu and Jiang Liu",

note = "Publisher Copyright: {\textcopyright} Springer International Publishing AG 2017.; International Workshop on Fetal and Infant Image Analysis, FIFI 2017 and 4th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2017 held in Conjunction with 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017 ; Conference date: 14-09-2017 Through 14-09-2017",

year = "2017",

doi = "10.1007/978-3-319-67561-9\_14",

language = "English",

isbn = "9783319675602",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "126--133",

editor = "Cardoso, \{M. Jorge\} and Tal Arbel",

booktitle = "Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings",

address = "Germany",

}

Sun, J, Wan, C, Cheng, J, Yu, F & Liu, J 2017, Retinal image quality classification using fine-tuned CNN. in MJ Cardoso & T Arbel (eds), Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10554 LNCS, Springer Verlag, pp. 126-133, International Workshop on Fetal and Infant Image Analysis, FIFI 2017 and 4th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2017 held in Conjunction with 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017, Quebec City, Canada, 14/09/17. https://doi.org/10.1007/978-3-319-67561-9_14

Retinal image quality classification using fine-tuned CNN. / Sun, Jing; Wan, Cheng; Cheng, Jun et al.
Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings. ed. / M. Jorge Cardoso; Tal Arbel. Springer Verlag, 2017. p. 126-133 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10554 LNCS).

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Retinal image quality classification using fine-tuned CNN

AU - Sun, Jing

AU - Wan, Cheng

AU - Cheng, Jun

AU - Yu, Fengli

AU - Liu, Jiang

N1 - Publisher Copyright: © Springer International Publishing AG 2017.

PY - 2017

Y1 - 2017

N2 - Retinal image quality classification makes a great difference in automated diabetic retinopathy screening systems. With the increase of application of portable fundus cameras, we can get a large number of retinal images, but there are quite a number of images in poor quality because of uneven illumination, occlusion and patients movements. Using the dataset with poor quality training networks for DR screening system will lead to the decrease of accuracy. In this paper, we first explore four CNN architectures (AlexNet, GoogLeNet, VGG-16, and ResNet-50) from ImageNet image classification task to our Retinal fundus images quality classification, then we pick top two networks out and jointly fine-tune the two networks. The total loss of the network we proposed is equal to the sum of the losses of all channels. We demonstrate the super performance of our proposed algorithm on a large retinal fundus image dataset and achieve an optimal accuracy of 97.12%, outperforming the current methods in this area.

AB - Retinal image quality classification makes a great difference in automated diabetic retinopathy screening systems. With the increase of application of portable fundus cameras, we can get a large number of retinal images, but there are quite a number of images in poor quality because of uneven illumination, occlusion and patients movements. Using the dataset with poor quality training networks for DR screening system will lead to the decrease of accuracy. In this paper, we first explore four CNN architectures (AlexNet, GoogLeNet, VGG-16, and ResNet-50) from ImageNet image classification task to our Retinal fundus images quality classification, then we pick top two networks out and jointly fine-tune the two networks. The total loss of the network we proposed is equal to the sum of the losses of all channels. We demonstrate the super performance of our proposed algorithm on a large retinal fundus image dataset and achieve an optimal accuracy of 97.12%, outperforming the current methods in this area.

KW - Convolutional neural networks (CNN)

KW - Fine-tuning

KW - No-reference image quality assessment (NR-IQA)

KW - Retinal image

UR - https://www.scopus.com/pages/publications/85029821249

U2 - 10.1007/978-3-319-67561-9_14

DO - 10.1007/978-3-319-67561-9_14

M3 - Conference contribution

AN - SCOPUS:85029821249

SN - 9783319675602

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 126

EP - 133

BT - Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings

A2 - Cardoso, M. Jorge

A2 - Arbel, Tal

PB - Springer Verlag

T2 - International Workshop on Fetal and Infant Image Analysis, FIFI 2017 and 4th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2017 held in Conjunction with 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017

Y2 - 14 September 2017 through 14 September 2017

ER -

Sun J, Wan C, Cheng J, Yu F, Liu J. Retinal image quality classification using fine-tuned CNN. In Cardoso MJ, Arbel T, editors, Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings. Springer Verlag. 2017. p. 126-133. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-319-67561-9_14

Retinal image quality classification using fine-tuned CNN

Abstract

Publication series

Conference

UN SDGs

Free Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this