Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks

Ling Ping Cen; Jie Ji; Jian Wei Lin; Si Tong Ju; Hong Jie Lin; Tai Ping Li; Yun Wang; Jian Feng Yang; Yu Fen Liu; Shaoying Tan; Li Tan; Dongjie Li; Yifan Wang; Dezhi Zheng; Yongqun Xiong; Hanfu Wu; Jingjing Jiang; Zhenggen Wu; Dingguo Huang; Tingkun Shi; Binyao Chen; Jianling Yang; Xiaoling Zhang; Li Luo; Chukai Huang; Guihua Zhang; Yuqiang Huang; Tsz Kin Ng; Haoyu Chen; Weiqi Chen; Chi Pui Pang; Mingzhi Zhang

doi:10.1038/s41467-021-25138-w

Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks

Ling Ping Cen, Jie Ji, Jian Wei Lin, Si Tong Ju, Hong Jie Lin, Tai Ping Li, Yun Wang, Jian Feng Yang, Yu Fen Liu, Shaoying Tan, Li Tan, Dongjie Li, Yifan Wang, Dezhi Zheng, Yongqun Xiong, Hanfu Wu, Jingjing Jiang, Zhenggen Wu, Dingguo Huang, Tingkun ShiBinyao Chen, Jianling Yang, Xiaoling Zhang, Li Luo, Chukai Huang, Guihua Zhang, Yuqiang Huang, Tsz Kin Ng, Haoyu Chen, Weiqi Chen, Chi Pui Pang, Mingzhi Zhang

Research output: Journal article publication › Journal article › Academic research › peer-review

137 Citations (Scopus)

Abstract

Retinal fundus diseases can lead to irreversible visual impairment without timely diagnoses and appropriate treatments. Single disease-based deep learning algorithms had been developed for the detection of diabetic retinopathy, age-related macular degeneration, and glaucoma. Here, we developed a deep learning platform (DLP) capable of detecting multiple common referable fundus diseases and conditions (39 classes) by using 249,620 fundus images marked with 275,543 labels from heterogenous sources. Our DLP achieved a frequency-weighted average F1 score of 0.923, sensitivity of 0.978, specificity of 0.996 and area under the receiver operating characteristic curve (AUC) of 0.9984 for multi-label classification in the primary test dataset and reached the average level of retina specialists. External multihospital test, public data test and tele-reading application also showed high efficiency for multiple retinal diseases and conditions detection. These results indicate that our DLP can be applied for retinal fundus disease triage, especially in remote areas around the world.

Original language	English
Article number	4828
Number of pages	13
Journal	Nature Communications
Volume	12
Issue number	1
DOIs	https://doi.org/10.1038/s41467-021-25138-w
Publication status	Published - 10 Aug 2021
Externally published	Yes

ASJC Scopus subject areas

General Chemistry
General Biochemistry,Genetics and Molecular Biology
General Physics and Astronomy

UN SDGs

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

More information

10.1038/s41467-021-25138-w

Cite this

Cen, L. P., Ji, J., Lin, J. W., Ju, S. T., Lin, H. J., Li, T. P., Wang, Y., Yang, J. F., Liu, Y. F., Tan, S., Tan, L., Li, D., Wang, Y., Zheng, D., Xiong, Y., Wu, H., Jiang, J., Wu, Z., Huang, D., ... Zhang, M. (2021). Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks. Nature Communications, 12(1), Article 4828. https://doi.org/10.1038/s41467-021-25138-w

@article{fb15c886b1cf42c68eeb2ce06e5d1625,

title = "Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks",

abstract = "Retinal fundus diseases can lead to irreversible visual impairment without timely diagnoses and appropriate treatments. Single disease-based deep learning algorithms had been developed for the detection of diabetic retinopathy, age-related macular degeneration, and glaucoma. Here, we developed a deep learning platform (DLP) capable of detecting multiple common referable fundus diseases and conditions (39 classes) by using 249,620 fundus images marked with 275,543 labels from heterogenous sources. Our DLP achieved a frequency-weighted average F1 score of 0.923, sensitivity of 0.978, specificity of 0.996 and area under the receiver operating characteristic curve (AUC) of 0.9984 for multi-label classification in the primary test dataset and reached the average level of retina specialists. External multihospital test, public data test and tele-reading application also showed high efficiency for multiple retinal diseases and conditions detection. These results indicate that our DLP can be applied for retinal fundus disease triage, especially in remote areas around the world.",

author = "Cen, {Ling Ping} and Jie Ji and Lin, {Jian Wei} and Ju, {Si Tong} and Lin, {Hong Jie} and Li, {Tai Ping} and Yun Wang and Yang, {Jian Feng} and Liu, {Yu Fen} and Shaoying Tan and Li Tan and Dongjie Li and Yifan Wang and Dezhi Zheng and Yongqun Xiong and Hanfu Wu and Jingjing Jiang and Zhenggen Wu and Dingguo Huang and Tingkun Shi and Binyao Chen and Jianling Yang and Xiaoling Zhang and Li Luo and Chukai Huang and Guihua Zhang and Yuqiang Huang and Ng, {Tsz Kin} and Haoyu Chen and Weiqi Chen and Pang, {Chi Pui} and Mingzhi Zhang",

note = "Funding Information: This study was supported by the National Natural Science Foundation of China (NSFC, 81570849 to L.-P.C., 81800822 to S.T.), Natural Science Foundation of Guangdong Province, China (NSFG, 2020A1515010415 to L.-P.C.), Grant for Key Disciplinary Project of Clinical Medicine under the Guangdong High-level University Development Program (002-18119101), LKSF cross-disciplinary research grants (2020LKSFG16B to L.-P.C. and M.Z.), and an internal grant from the Joint Shantou International Eye Center of The Shantou University and The Chinese University of Hong Kong (17-003 to L.-P.C.). Publisher Copyright: {\textcopyright} 2021, The Author(s).",

year = "2021",

month = aug,

day = "10",

doi = "10.1038/s41467-021-25138-w",

language = "English",

volume = "12",

journal = "Nature Communications",

issn = "2041-1723",

publisher = "Nature Publishing Group",

number = "1",

}

Cen, LP, Ji, J, Lin, JW, Ju, ST, Lin, HJ, Li, TP, Wang, Y, Yang, JF, Liu, YF, Tan, S, Tan, L, Li, D, Wang, Y, Zheng, D, Xiong, Y, Wu, H, Jiang, J, Wu, Z, Huang, D, Shi, T, Chen, B, Yang, J, Zhang, X, Luo, L, Huang, C, Zhang, G, Huang, Y, Ng, TK, Chen, H, Chen, W, Pang, CP & Zhang, M 2021, 'Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks', Nature Communications, vol. 12, no. 1, 4828. https://doi.org/10.1038/s41467-021-25138-w

TY - JOUR

T1 - Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks

AU - Cen, Ling Ping

AU - Ji, Jie

AU - Lin, Jian Wei

AU - Ju, Si Tong

AU - Lin, Hong Jie

AU - Li, Tai Ping

AU - Wang, Yun

AU - Yang, Jian Feng

AU - Liu, Yu Fen

AU - Tan, Shaoying

AU - Tan, Li

AU - Li, Dongjie

AU - Wang, Yifan

AU - Zheng, Dezhi

AU - Xiong, Yongqun

AU - Wu, Hanfu

AU - Jiang, Jingjing

AU - Wu, Zhenggen

AU - Huang, Dingguo

AU - Shi, Tingkun

AU - Chen, Binyao

AU - Yang, Jianling

AU - Zhang, Xiaoling

AU - Luo, Li

AU - Huang, Chukai

AU - Zhang, Guihua

AU - Huang, Yuqiang

AU - Ng, Tsz Kin

AU - Chen, Haoyu

AU - Chen, Weiqi

AU - Pang, Chi Pui

AU - Zhang, Mingzhi

N1 - Funding Information: This study was supported by the National Natural Science Foundation of China (NSFC, 81570849 to L.-P.C., 81800822 to S.T.), Natural Science Foundation of Guangdong Province, China (NSFG, 2020A1515010415 to L.-P.C.), Grant for Key Disciplinary Project of Clinical Medicine under the Guangdong High-level University Development Program (002-18119101), LKSF cross-disciplinary research grants (2020LKSFG16B to L.-P.C. and M.Z.), and an internal grant from the Joint Shantou International Eye Center of The Shantou University and The Chinese University of Hong Kong (17-003 to L.-P.C.). Publisher Copyright: © 2021, The Author(s).

PY - 2021/8/10

Y1 - 2021/8/10

N2 - Retinal fundus diseases can lead to irreversible visual impairment without timely diagnoses and appropriate treatments. Single disease-based deep learning algorithms had been developed for the detection of diabetic retinopathy, age-related macular degeneration, and glaucoma. Here, we developed a deep learning platform (DLP) capable of detecting multiple common referable fundus diseases and conditions (39 classes) by using 249,620 fundus images marked with 275,543 labels from heterogenous sources. Our DLP achieved a frequency-weighted average F1 score of 0.923, sensitivity of 0.978, specificity of 0.996 and area under the receiver operating characteristic curve (AUC) of 0.9984 for multi-label classification in the primary test dataset and reached the average level of retina specialists. External multihospital test, public data test and tele-reading application also showed high efficiency for multiple retinal diseases and conditions detection. These results indicate that our DLP can be applied for retinal fundus disease triage, especially in remote areas around the world.

AB - Retinal fundus diseases can lead to irreversible visual impairment without timely diagnoses and appropriate treatments. Single disease-based deep learning algorithms had been developed for the detection of diabetic retinopathy, age-related macular degeneration, and glaucoma. Here, we developed a deep learning platform (DLP) capable of detecting multiple common referable fundus diseases and conditions (39 classes) by using 249,620 fundus images marked with 275,543 labels from heterogenous sources. Our DLP achieved a frequency-weighted average F1 score of 0.923, sensitivity of 0.978, specificity of 0.996 and area under the receiver operating characteristic curve (AUC) of 0.9984 for multi-label classification in the primary test dataset and reached the average level of retina specialists. External multihospital test, public data test and tele-reading application also showed high efficiency for multiple retinal diseases and conditions detection. These results indicate that our DLP can be applied for retinal fundus disease triage, especially in remote areas around the world.

UR - http://www.scopus.com/inward/record.url?scp=85112046519&partnerID=8YFLogxK

U2 - 10.1038/s41467-021-25138-w

DO - 10.1038/s41467-021-25138-w

M3 - Journal article

C2 - 34376678

AN - SCOPUS:85112046519

SN - 2041-1723

VL - 12

JO - Nature Communications

JF - Nature Communications

IS - 1

M1 - 4828

ER -

Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks

Abstract

ASJC Scopus subject areas

UN SDGs

More information

Other files and links

Fingerprint

Cite this