Wu Yufei, Jiang Jiahui, Deng Xiaoyu, Zhang Xixi, Lu Jinger, Xu Zian, Zhao Yitian, Chi Zai-Long, Lu Qinkang
Ophthalmology Center, The Affiliated Peoples Hospital of Ningbo University, Ningbo, Zhejiang, China.
School of Ophthalmology and Optometry, Wenzhou Medical University, Wenzhou, Zhejiang, China.
Front Cell Dev Biol. 2025 May 26;13:1609928. doi: 10.3389/fcell.2025.1609928. eCollection 2025.
To evaluate and quantify diabetes-related retinal and choroid perfusion changes in individuals with and without high myopia and explore their associations with diabetes risk factors.
Diabetic patients [n = 133; 43 without diabetic retinopathy in group DM; 48 non-proliferative diabetic retinopathies in group DR; 42 without DR but with high myopia in group HM] underwent ophthalmological and endocrinological examinations. Swept-source optical coherence tomography angiography (SS-OCTA) was used to image the retinal vessel density (RVD), retinal thickness (RT), choroidal thickness (CT), choriocapillaris vessel perfusion (CPV) and choroidal vascularity index (CVI). Automatic segmentation of retinal and choroidal layers was performed using a deep learning-based U-Net architecture. A ResNet-50 convolutional neural network was further applied to analyze vascular density patterns and assist in DR grading. Univariate and multiple linear regression analyses explored the associations between perfusion and risk factors.
The inner ring retinal vessel density and CVI in all areas were significantly different between groups ( < 0.05); CPV was not significantly changed except for the inferotemporal area among the groups. CT was decreased in all areas between groups ( < 0.05). The visual impairments in HM group was more obvious correlation with the retinal and choroidal structural changes. The AI-driven analysis revealed that decreased CVI and CT were significantly associated with age and spherical equivalent (SE), highlighting the utility of automated algorithms in identifying early microvascular impairments.
Diabetic patients with high myopia exhibited significantly lower CVI compared to those with diabetic retinopathy, indicating that CVI monitoring could facilitate risk stratification of diabetic retinopathy progression. The integration of SS-OCTA with artificial intelligence-enhanced segmentation and vascular analysis provides a refined method for early detection of retinal and choroidal microvascular impairments in diabetic populations.
评估和量化有无高度近视个体中与糖尿病相关的视网膜和脉络膜灌注变化，并探讨它们与糖尿病风险因素的关联。
糖尿病患者[共133例；糖尿病组（DM）中43例无糖尿病视网膜病变；糖尿病视网膜病变组（DR）中48例非增殖性糖尿病视网膜病变；高度近视组（HM）中42例无糖尿病视网膜病变但有高度近视]接受了眼科和内分泌检查。使用扫频光学相干断层扫描血管造影（SS-OCTA）对视网膜血管密度（RVD）、视网膜厚度（RT）、脉络膜厚度（CT）、脉络膜毛细血管灌注（CPV）和脉络膜血管指数（CVI）进行成像。使用基于深度学习的U-Net架构对视网膜和脉络膜层进行自动分割。进一步应用ResNet-50卷积神经网络分析血管密度模式并辅助糖尿病视网膜病变分级。单因素和多因素线性回归分析探讨灌注与风险因素之间的关联。
各组之间所有区域的内环视网膜血管密度和CVI均有显著差异（<0.05）；除颞下区域外，各组之间CPV无显著变化。各组之间所有区域的CT均降低（<0.05）。高度近视组的视力损害与视网膜和脉络膜结构变化的相关性更明显。人工智能驱动的分析显示，CVI和CT降低与年龄和等效球镜度（SE）显著相关，突出了自动算法在识别早期微血管损伤方面的实用性。
与糖尿病视网膜病变患者相比，高度近视糖尿病患者的CVI显著更低，这表明监测CVI有助于糖尿病视网膜病变进展的风险分层。SS-OCTA与人工智能增强分割和血管分析相结合，为早期检测糖尿病患者的视网膜和脉络膜微血管损伤提供了一种精细的方法。
