A Deep Learning Segmentation Model for Detection of Active Proliferative Diabetic Retinopathy

Abstract Introduction Existing deep learning (DL) algorithms lack the capability to accurately identify patients in immediate need of treatment for proliferative diabetic retinopathy (PDR). We aimed to develop a DL segmentation model to detect active PDR in six-field retinal images by the annotation...

Full description

Saved in:
Bibliographic Details
Main Authors: Sebastian Dinesen, Marianne G. Schou, Christoffer V. Hedegaard, Yousif Subhi, Thiusius R. Savarimuthu, Tunde Peto, Jakob K. H. Andersen, Jakob Grauslund
Format: Article
Language:English
Published: Adis, Springer Healthcare 2025-03-01
Series:Ophthalmology and Therapy
Subjects:
Online Access:https://doi.org/10.1007/s40123-025-01127-w
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849699745141882880
author Sebastian Dinesen
Marianne G. Schou
Christoffer V. Hedegaard
Yousif Subhi
Thiusius R. Savarimuthu
Tunde Peto
Jakob K. H. Andersen
Jakob Grauslund
author_facet Sebastian Dinesen
Marianne G. Schou
Christoffer V. Hedegaard
Yousif Subhi
Thiusius R. Savarimuthu
Tunde Peto
Jakob K. H. Andersen
Jakob Grauslund
author_sort Sebastian Dinesen
collection DOAJ
description Abstract Introduction Existing deep learning (DL) algorithms lack the capability to accurately identify patients in immediate need of treatment for proliferative diabetic retinopathy (PDR). We aimed to develop a DL segmentation model to detect active PDR in six-field retinal images by the annotation of new retinal vessels and preretinal hemorrhages. Methods We identified six-field retinal images classified at level 4 of the International Clinical Diabetic Retinopathy Disease Severity Scale collected at the Island of Funen from 2009 to 2019 as part of the Danish screening program for diabetic retinopathy (DR). A certified grader (grader 1) manually dichotomized the images into active or inactive PDR, and the images were then reassessed by two independent certified graders. In cases of disagreement, the final classification decision was made in collaboration between grader 1 and one of the secondary graders. Overall, 637 images were classified as active PDR. We then applied our pre-established DL segmentation model to annotate nine lesion types before training the algorithm. The segmentations of new vessels and preretinal hemorrhages were corrected for any inaccuracies before training the DL algorithm. After the classification and pre-segmentation phases the images were divided into training (70%), validation (10%), and testing (20%) datasets. We added 301 images with inactive PDR to the testing dataset. Results We included 637 images of active PDR and 301 images of inactive PDR from 199 individuals. The training dataset had 1381 new vessel and preretinal hemorrhage lesions, while the validation dataset had 123 lesions and the testing dataset 374 lesions. The DL system demonstrated a sensitivity of 90% and a specificity of 70% for annotation-assisted classification of active PDR. The negative predictive value was 94%, while the positive predictive value was 57%. Conclusions Our DL segmentation model achieved excellent sensitivity and acceptable specificity in distinguishing active from inactive PDR.
format Article
id doaj-art-c6947a5e55724110b3b0e200d2192fc4
institution DOAJ
issn 2193-8245
2193-6528
language English
publishDate 2025-03-01
publisher Adis, Springer Healthcare
record_format Article
series Ophthalmology and Therapy
spelling doaj-art-c6947a5e55724110b3b0e200d2192fc42025-08-20T03:18:30ZengAdis, Springer HealthcareOphthalmology and Therapy2193-82452193-65282025-03-011451053106310.1007/s40123-025-01127-wA Deep Learning Segmentation Model for Detection of Active Proliferative Diabetic RetinopathySebastian Dinesen0Marianne G. Schou1Christoffer V. Hedegaard2Yousif Subhi3Thiusius R. Savarimuthu4Tunde Peto5Jakob K. H. Andersen6Jakob Grauslund7Department of Ophthalmology, Odense University HospitalDepartment of Ophthalmology, Odense University HospitalDepartment of Ophthalmology, Odense University HospitalDepartment of Clinical Research, University of Southern DenmarkThe Maersk Mc-Kinney Moller Institute, University of Southern DenmarkCentre for Public Health, Queen’s University BelfastThe Maersk Mc-Kinney Moller Institute, University of Southern DenmarkDepartment of Ophthalmology, Odense University HospitalAbstract Introduction Existing deep learning (DL) algorithms lack the capability to accurately identify patients in immediate need of treatment for proliferative diabetic retinopathy (PDR). We aimed to develop a DL segmentation model to detect active PDR in six-field retinal images by the annotation of new retinal vessels and preretinal hemorrhages. Methods We identified six-field retinal images classified at level 4 of the International Clinical Diabetic Retinopathy Disease Severity Scale collected at the Island of Funen from 2009 to 2019 as part of the Danish screening program for diabetic retinopathy (DR). A certified grader (grader 1) manually dichotomized the images into active or inactive PDR, and the images were then reassessed by two independent certified graders. In cases of disagreement, the final classification decision was made in collaboration between grader 1 and one of the secondary graders. Overall, 637 images were classified as active PDR. We then applied our pre-established DL segmentation model to annotate nine lesion types before training the algorithm. The segmentations of new vessels and preretinal hemorrhages were corrected for any inaccuracies before training the DL algorithm. After the classification and pre-segmentation phases the images were divided into training (70%), validation (10%), and testing (20%) datasets. We added 301 images with inactive PDR to the testing dataset. Results We included 637 images of active PDR and 301 images of inactive PDR from 199 individuals. The training dataset had 1381 new vessel and preretinal hemorrhage lesions, while the validation dataset had 123 lesions and the testing dataset 374 lesions. The DL system demonstrated a sensitivity of 90% and a specificity of 70% for annotation-assisted classification of active PDR. The negative predictive value was 94%, while the positive predictive value was 57%. Conclusions Our DL segmentation model achieved excellent sensitivity and acceptable specificity in distinguishing active from inactive PDR.https://doi.org/10.1007/s40123-025-01127-wDeep learningProliferative diabetic retinopathyBlack box eliminationNew vesselsPreretinal hemorrhages
spellingShingle Sebastian Dinesen
Marianne G. Schou
Christoffer V. Hedegaard
Yousif Subhi
Thiusius R. Savarimuthu
Tunde Peto
Jakob K. H. Andersen
Jakob Grauslund
A Deep Learning Segmentation Model for Detection of Active Proliferative Diabetic Retinopathy
Ophthalmology and Therapy
Deep learning
Proliferative diabetic retinopathy
Black box elimination
New vessels
Preretinal hemorrhages
title A Deep Learning Segmentation Model for Detection of Active Proliferative Diabetic Retinopathy
title_full A Deep Learning Segmentation Model for Detection of Active Proliferative Diabetic Retinopathy
title_fullStr A Deep Learning Segmentation Model for Detection of Active Proliferative Diabetic Retinopathy
title_full_unstemmed A Deep Learning Segmentation Model for Detection of Active Proliferative Diabetic Retinopathy
title_short A Deep Learning Segmentation Model for Detection of Active Proliferative Diabetic Retinopathy
title_sort deep learning segmentation model for detection of active proliferative diabetic retinopathy
topic Deep learning
Proliferative diabetic retinopathy
Black box elimination
New vessels
Preretinal hemorrhages
url https://doi.org/10.1007/s40123-025-01127-w
work_keys_str_mv AT sebastiandinesen adeeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT mariannegschou adeeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT christoffervhedegaard adeeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT yousifsubhi adeeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT thiusiusrsavarimuthu adeeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT tundepeto adeeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT jakobkhandersen adeeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT jakobgrauslund adeeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT sebastiandinesen deeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT mariannegschou deeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT christoffervhedegaard deeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT yousifsubhi deeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT thiusiusrsavarimuthu deeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT tundepeto deeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT jakobkhandersen deeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy
AT jakobgrauslund deeplearningsegmentationmodelfordetectionofactiveproliferativediabeticretinopathy