A deep semi-supervised learning approach to the detection of glaucoma on out-of-distribution retinal fundus image datasets

Abstract Background Accurate detection of glaucoma plays a critical role in treating the disease and can be performed on limited labeled retinal fundus images and large-scale unlabeled ones leveraging a deep semi-supervised learning (SSL) technology. This study aims to investigate how glaucoma depic...

Full description

Saved in:
Bibliographic Details
Main Authors: Lei Wang, Xiaoyun Zhang, Zhongwen Li, Shuchen Yu, Yabo Wu, Shaodan Zhang, Gaoqiang Jiang, Bihan Tian, Chenyang Mei, Jiantao Pu, Yuanbo Liang, Quanyong Yi, Wencan Wu
Format: Article
Language:English
Published: BMC 2025-05-01
Series:BMC Ophthalmology
Subjects:
Online Access:https://doi.org/10.1186/s12886-025-04153-1
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849688062254120960
author Lei Wang
Xiaoyun Zhang
Zhongwen Li
Shuchen Yu
Yabo Wu
Shaodan Zhang
Gaoqiang Jiang
Bihan Tian
Chenyang Mei
Jiantao Pu
Yuanbo Liang
Quanyong Yi
Wencan Wu
author_facet Lei Wang
Xiaoyun Zhang
Zhongwen Li
Shuchen Yu
Yabo Wu
Shaodan Zhang
Gaoqiang Jiang
Bihan Tian
Chenyang Mei
Jiantao Pu
Yuanbo Liang
Quanyong Yi
Wencan Wu
author_sort Lei Wang
collection DOAJ
description Abstract Background Accurate detection of glaucoma plays a critical role in treating the disease and can be performed on limited labeled retinal fundus images and large-scale unlabeled ones leveraging a deep semi-supervised learning (SSL) technology. This study aims to investigate how glaucoma depicted on fundus images can be reliably detected by the SSL technology and the impact of the quantities and qualities of unlabeled images on the outcome. Methods We retrospectively collected a dataset consisting of 7,503 fundus images and classified them into four categories, namely none, mild, moderate, or severe glaucoma. We used the collected dataset and a public out-of-distribution (OOD) dataset (EyeQ) to train an available SSL method (called SRC-MT) to grade glaucoma. Results SRC-MT achieved an average area under the receiver operating characteristic curve (AUC) of 0.8944 and 0.8969 on global field-of-view (FOV) regions and local disc regions, respectively when trained on 600 labeled images and 5401 unlabeled ones from the collected dataset. When separately introducing 16,817, 6,435, and 5,540 unlabeled OOD images with the qualities of ‘good’, ‘usable’, and ‘reject’ from the EyeQ dataset into 5,401 unlabeled images, its performance became 0.8972, 0.8908, and 0.8922, respectively for global FOV regions on the testing subset from the collected dataset, and 0.7342, 0.5090, and 0.5072 on three public datasets (i.e., AIROGS, EDDFS, and FIVES). Conclusions SRC-MT achieved promising performance for glaucoma grading, especially in global FOV regions. Its performance increased when using more labeled images, but degraded when using more unlabeled OOD images with worse image qualities.
format Article
id doaj-art-11bc6c6cab7e4d0c8f644d0b23308375
institution DOAJ
issn 1471-2415
language English
publishDate 2025-05-01
publisher BMC
record_format Article
series BMC Ophthalmology
spelling doaj-art-11bc6c6cab7e4d0c8f644d0b233083752025-08-20T03:22:08ZengBMCBMC Ophthalmology1471-24152025-05-0125111710.1186/s12886-025-04153-1A deep semi-supervised learning approach to the detection of glaucoma on out-of-distribution retinal fundus image datasetsLei Wang0Xiaoyun Zhang1Zhongwen Li2Shuchen Yu3Yabo Wu4Shaodan Zhang5Gaoqiang Jiang6Bihan Tian7Chenyang Mei8Jiantao Pu9Yuanbo Liang10Quanyong Yi11Wencan Wu12National Engineering Research Center of Ophthalmology and Optometry, Eye Hospital, Wenzhou Medical UniversityThe University of SydneyThe Affiliated Ningbo Eye Hospital of Wenzhou Medical UniversityNational Engineering Research Center of Ophthalmology and Optometry, Eye Hospital, Wenzhou Medical UniversitySchool of Biomedical Engineering and Imaging Sciences, King’s college LondonNational Engineering Research Center of Ophthalmology and Optometry, Eye Hospital, Wenzhou Medical UniversityNational Engineering Research Center of Ophthalmology and Optometry, Eye Hospital, Wenzhou Medical UniversityNational Engineering Research Center of Ophthalmology and Optometry, Eye Hospital, Wenzhou Medical UniversityNational Engineering Research Center of Ophthalmology and Optometry, Eye Hospital, Wenzhou Medical UniversityDepartments of Radiology and Bioengineering, University of PittsburghNational Engineering Research Center of Ophthalmology and Optometry, Eye Hospital, Wenzhou Medical UniversityThe Affiliated Ningbo Eye Hospital of Wenzhou Medical UniversityNational Engineering Research Center of Ophthalmology and Optometry, Eye Hospital, Wenzhou Medical UniversityAbstract Background Accurate detection of glaucoma plays a critical role in treating the disease and can be performed on limited labeled retinal fundus images and large-scale unlabeled ones leveraging a deep semi-supervised learning (SSL) technology. This study aims to investigate how glaucoma depicted on fundus images can be reliably detected by the SSL technology and the impact of the quantities and qualities of unlabeled images on the outcome. Methods We retrospectively collected a dataset consisting of 7,503 fundus images and classified them into four categories, namely none, mild, moderate, or severe glaucoma. We used the collected dataset and a public out-of-distribution (OOD) dataset (EyeQ) to train an available SSL method (called SRC-MT) to grade glaucoma. Results SRC-MT achieved an average area under the receiver operating characteristic curve (AUC) of 0.8944 and 0.8969 on global field-of-view (FOV) regions and local disc regions, respectively when trained on 600 labeled images and 5401 unlabeled ones from the collected dataset. When separately introducing 16,817, 6,435, and 5,540 unlabeled OOD images with the qualities of ‘good’, ‘usable’, and ‘reject’ from the EyeQ dataset into 5,401 unlabeled images, its performance became 0.8972, 0.8908, and 0.8922, respectively for global FOV regions on the testing subset from the collected dataset, and 0.7342, 0.5090, and 0.5072 on three public datasets (i.e., AIROGS, EDDFS, and FIVES). Conclusions SRC-MT achieved promising performance for glaucoma grading, especially in global FOV regions. Its performance increased when using more labeled images, but degraded when using more unlabeled OOD images with worse image qualities.https://doi.org/10.1186/s12886-025-04153-1Glaucoma detectionFundus imagesSemi-supervised learningOut-of-distributionQuantities and qualities of images
spellingShingle Lei Wang
Xiaoyun Zhang
Zhongwen Li
Shuchen Yu
Yabo Wu
Shaodan Zhang
Gaoqiang Jiang
Bihan Tian
Chenyang Mei
Jiantao Pu
Yuanbo Liang
Quanyong Yi
Wencan Wu
A deep semi-supervised learning approach to the detection of glaucoma on out-of-distribution retinal fundus image datasets
BMC Ophthalmology
Glaucoma detection
Fundus images
Semi-supervised learning
Out-of-distribution
Quantities and qualities of images
title A deep semi-supervised learning approach to the detection of glaucoma on out-of-distribution retinal fundus image datasets
title_full A deep semi-supervised learning approach to the detection of glaucoma on out-of-distribution retinal fundus image datasets
title_fullStr A deep semi-supervised learning approach to the detection of glaucoma on out-of-distribution retinal fundus image datasets
title_full_unstemmed A deep semi-supervised learning approach to the detection of glaucoma on out-of-distribution retinal fundus image datasets
title_short A deep semi-supervised learning approach to the detection of glaucoma on out-of-distribution retinal fundus image datasets
title_sort deep semi supervised learning approach to the detection of glaucoma on out of distribution retinal fundus image datasets
topic Glaucoma detection
Fundus images
Semi-supervised learning
Out-of-distribution
Quantities and qualities of images
url https://doi.org/10.1186/s12886-025-04153-1
work_keys_str_mv AT leiwang adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT xiaoyunzhang adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT zhongwenli adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT shuchenyu adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT yabowu adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT shaodanzhang adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT gaoqiangjiang adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT bihantian adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT chenyangmei adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT jiantaopu adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT yuanboliang adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT quanyongyi adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT wencanwu adeepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT leiwang deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT xiaoyunzhang deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT zhongwenli deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT shuchenyu deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT yabowu deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT shaodanzhang deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT gaoqiangjiang deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT bihantian deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT chenyangmei deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT jiantaopu deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT yuanboliang deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT quanyongyi deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets
AT wencanwu deepsemisupervisedlearningapproachtothedetectionofglaucomaonoutofdistributionretinalfundusimagedatasets