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

IF 2.6 3区医学 Q2 OPHTHALMOLOGY

Ophthalmology and Therapy Pub Date : 2025-03-27 DOI:10.1007/s40123-025-01127-w

Sebastian Dinesen, Marianne G Schou, Christoffer V Hedegaard, Yousif Subhi, Thiusius R Savarimuthu, Tunde Peto, Jakob K H Andersen, Jakob Grauslund

{"title":"A Deep Learning Segmentation Model for Detection of Active Proliferative Diabetic Retinopathy.","authors":"Sebastian Dinesen, Marianne G Schou, Christoffer V Hedegaard, Yousif Subhi, Thiusius R Savarimuthu, Tunde Peto, Jakob K H Andersen, Jakob Grauslund","doi":"10.1007/s40123-025-01127-w","DOIUrl":null,"url":null,"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.","PeriodicalId":19623,"journal":{"name":"Ophthalmology and Therapy","volume":" ","pages":""},"PeriodicalIF":2.6000,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Ophthalmology and Therapy","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1007/s40123-025-01127-w","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"OPHTHALMOLOGY","Score":null,"Total":0}

引用次数: 0

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.

查看原文本刊更多论文

活动性增殖性糖尿病视网膜病变检测的深度学习分割模型。

现有的深度学习（DL）算法缺乏准确识别急需治疗的增殖性糖尿病视网膜病变（PDR）患者的能力。我们的目的是建立一个DL分割模型，通过注释新的视网膜血管和视网膜前出血来检测六视场视网膜图像中的活性PDR。方法：作为丹麦糖尿病视网膜病变（DR）筛查计划的一部分，我们确定了2009年至2019年在Funen岛收集的国际临床糖尿病视网膜病变疾病严重程度量表中分类为4级的六场视网膜图像。经过认证的评分员（评分员1）手动将图像分为活动或非活动PDR，然后由两个独立的经过认证的评分员重新评估图像。在意见不一致的情况下，最终的分类决定由一年级学生和一名二年级学生共同决定。总共有637张图像被归类为活动PDR。然后，在训练算法之前，我们应用预先建立的深度学习分割模型对9种病变类型进行标注。在训练DL算法之前，对新血管和视网膜前出血的分割进行了任何不准确的校正。经过分类和预分割阶段，将图像分为训练（70%）、验证（10%）和测试（20%）数据集。我们向测试数据集添加了301张具有非活动PDR的图像。结果：共纳入199个个体的活性PDR图像637张，非活性PDR图像301张。训练数据集中有1381个新的血管和视网膜前出血病变，而验证数据集中有123个病变，测试数据集中有374个病变。DL系统对活性PDR的注释辅助分类的灵敏度为90%，特异性为70%。阴性预测值为94%，阳性预测值为57%。结论：我们的DL分割模型在区分活性和非活性PDR方面具有良好的敏感性和可接受的特异性。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Ophthalmology and Therapy OPHTHALMOLOGY-

CiteScore

4.20

自引率

3.00%

发文量

157

审稿时长

6 weeks

期刊介绍： Aims and Scope Ophthalmology and Therapy is an international, open access, peer-reviewed (single-blind), and rapid publication journal. The scope of the journal is broad and will consider all scientifically sound research from preclinical, clinical (all phases), observational, real-world, and health outcomes research around the use of ophthalmological therapies, devices, and surgical techniques. The journal is of interest to a broad audience of pharmaceutical and healthcare professionals and publishes original research, reviews, case reports/series, trial protocols and short communications such as commentaries and editorials. Ophthalmology and Therapy will consider all scientifically sound research be it positive, confirmatory or negative data. Submissions are welcomed whether they relate to an international and/or a country-specific audience, something that is crucially important when researchers are trying to target more specific patient populations. This inclusive approach allows the journal to assist in the dissemination of quality research, which may be considered of insufficient interest by other journals. Rapid Publication The journal’s publication timelines aim for a rapid peer review of 2 weeks. If an article is accepted it will be published 3–4 weeks from acceptance. The rapid timelines are achieved through the combination of a dedicated in-house editorial team, who manage article workflow, and an extensive Editorial and Advisory Board who assist with peer review. This allows the journal to support the rapid dissemination of research, whilst still providing robust peer review. Combined with the journal’s open access model this allows for the rapid, efficient communication of the latest research and reviews, fostering the advancement of ophthalmic therapies. Open Access All articles published by Ophthalmology and Therapy are open access. Personal Service The journal’s dedicated in-house editorial team offer a personal “concierge service” meaning authors will always have an editorial contact able to update them on the status of their manuscript. The editorial team check all manuscripts to ensure that articles conform to the most recent COPE, GPP and ICMJE publishing guidelines. This supports the publication of ethically sound and transparent research. Digital Features and Plain Language Summaries Ophthalmology and Therapy offers a range of additional features designed to increase the visibility, readership and educational value of the journal’s content. Each article is accompanied by key summary points, giving a time-efficient overview of the content to a wide readership. Articles may be accompanied by plain language summaries to assist readers who have some knowledge of, but not in-depth expertise in, the area to understand the scientific content and overall implications of the article. The journal also provides the option to include various types of digital features including animated abstracts, video abstracts, slide decks, audio slides, instructional videos, infographics, podcasts and animations. All additional features are peer reviewed to the same high standard as the article itself. If you consider that your paper would benefit from the inclusion of a digital feature, please let us know. Our editorial team are able to create high-quality slide decks and infographics in-house, and video abstracts through our partner Research Square, and would be happy to assist in any way we can. For further information about digital features, please contact the journal editor (see ‘Contact the Journal’ for email address), and see the ‘Guidelines for digital features and plain language summaries’ document under ‘Submission guidelines’. For examples of digital features please visit our showcase page https://springerhealthcare.com/expertise/publishing-digital-features/ Publication Fees Upon acceptance of an article, authors will be required to pay the mandatory Rapid Service Fee of €5250/$6000/£4300. The journal will consider fee discounts and waivers for developing countries and this is decided on a case by case basis. Peer Review Process Upon submission, manuscripts are assessed by the editorial team to ensure they fit within the aims and scope of the journal and are also checked for plagiarism. All suitable submissions are then subject to a comprehensive single-blind peer review. Reviewers are selected based on their relevant expertise and publication history in the subject area. The journal has an extensive pool of editorial and advisory board members who have been selected to assist with peer review based on the afore-mentioned criteria. At least two extensive reviews are required to make the editorial decision, with the exception of some article types such as Commentaries, Editorials, and Letters which are generally reviewed by one member of the Editorial Board. Where reviewer recommendations are conflicted, the editorial board will be contacted for further advice and a presiding decision. Manuscripts are then either accepted, rejected or authors are required to make major or minor revisions (both reviewer comments and editorial comments may need to be addressed). Once a revised manuscript is re-submitted, it is assessed along with the responses to reviewer comments and if it has been adequately revised it will be accepted for publication. Accepted manuscripts are then copyedited and typeset by the production team before online publication. Appeals against decisions following peer review are considered on a case-by-case basis and should be sent to the journal editor. Preprints We encourage posting of preprints of primary research manuscripts on preprint servers, authors’ or institutional websites, and open communications between researchers whether on community preprint servers or preprint commenting platforms. Posting of preprints is not considered prior publication and will not jeopardize consideration in our journals. Authors should disclose details of preprint posting during the submission process or at any other point during consideration in one of our journals. Once the manuscript is published, it is the author’s responsibility to ensure that the preprint record is updated with a publication reference, including the DOI and a URL link to the published version of the article on the journal website. Please follow the link for further information on preprint sharing: https://www.springer.com/gp/authors-editors/journal-author/journal-author-helpdesk/submission/1302#c16721550 Copyright Ophthalmology and Therapy''s content is published open access under the Creative Commons Attribution-Noncommercial License, which allows users to read, copy, distribute, and make derivative works for non-commercial purposes from the material, as long as the author of the original work is cited. The author assigns the exclusive right to any commercial use of the article to Springer. For more information about the Creative Commons Attribution-Noncommercial License, click here: http://creativecommons.org/licenses/by-nc/4.0. Contact For more information about the journal, including pre-submission enquiries, please contact christopher.vautrinot@springer.com.