Dinesen Sebastian, Schou Marianne G, Hedegaard Christoffer V, Subhi Yousif, Savarimuthu Thiusius R, Peto Tunde, Andersen Jakob K H, Grauslund Jakob
Department of Ophthalmology, Odense University Hospital, Sdr. Boulevard 29, 5000, Odense, Denmark.
Department of Clinical Research, University of Southern Denmark, Odense, Denmark.
Ophthalmol Ther. 2025 May;14(5):1053-1063. doi: 10.1007/s40123-025-01127-w. Epub 2025 Mar 27.
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.
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.
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%.
Our DL segmentation model achieved excellent sensitivity and acceptable specificity in distinguishing active from inactive PDR.
现有的深度学习（DL）算法缺乏准确识别急需增殖性糖尿病视网膜病变（PDR）治疗患者的能力。我们旨在开发一种DL分割模型，通过标注新的视网膜血管和视网膜前出血来检测六视野视网膜图像中的活动性PDR。
我们将2009年至2019年在菲英岛收集的国际临床糖尿病视网膜病变疾病严重程度分级为4级的六视野视网膜图像确定为丹麦糖尿病视网膜病变（DR）筛查项目的一部分。一名经过认证的分级人员（分级人员1）将图像手动分为活动性或非活动性PDR，然后由两名独立的经过认证的分级人员重新评估这些图像。在存在分歧的情况下，最终分类决定由分级人员1和一名二级分级人员共同做出。总体而言，637张图像被分类为活动性PDR。然后，在训练算法之前，我们应用预先建立的DL分割模型对九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之处的校正。在分类和预分割阶段之后，图像被分为训练（70%）、验证（10%）和测试（20%）数据集。我们将301张非活动性PDR图像添加到测试数据集中。
我们纳入了来自199名个体的637张活动性PDR图像和301张非活动性PDR图像。训练数据集有1381个新血管和视网膜前出血病变，验证数据集有123个病变，测试数据集有374个病变。DL系统在活动性PDR的标注辅助分类中显示出90%的敏感性和70%的特异性。阴性预测值为94%，而阳性预测值为57%。
我们的DL分割模型在区分活动性和非活动性PDR方面具有出色的敏感性和可接受的特异性。
