军事知识服务平台
基于人工智能眼底定量技术量化评估黄斑前膜视网膜血管特征
Quantitative assessment of retinal vascular morphological parameter analyses in eyes with idiopathic epiretinal membrane using artificial intelligence
摘要 本研究使用基于人工智能的EVisionAI眼底图像分析系统分析特发性黄斑前膜(ERM)患眼视网膜血管形态特征,回顾性分析了北京大学人民医院2023年1月~2025年3月的146名ERM患者292只眼的眼底彩照视网膜血管特征.根据患眼光相干断层眼底扫描检查记录患眼基于连续的异位中心凹内层结构(EIFL)的ERM分期、EIFL厚度及黄斑中心凹视网膜(CFT)厚度.统计分析视网膜血管形态特征参数的不同ERM分期的组间差异及其与ERM分期的相关性.结果表明,单侧ERM患者中,ERM眼较对侧健康眼血管平均分支夹角、血管平均管径、血管平均弯曲度更大,血管分形维数、血管弓和黄斑夹角更小(P<0.05);血管分形维数(P=0.004)、血管平均管径(P<0.001)、血管平均弯曲度(P<0.001)与ERM分期相关.血管分形维数<1.553(OR=7.247,P<0.001)、血管平均管径>61.93μm(OR=2.277,P=0.013)、血管平均弯曲度>0.893×104(OR=5.044,P<0.001)、血管弓和黄斑夹角<88.84°(OR=4.628,P<0.001)是ERM患眼达到ERM 3或4期的危险因素.本研究探索了各项视网膜血管参数与ERM分期的相关性及更高ERM分期的危险因素,为利用眼底彩照进行ERM分级、严重程度提供了一种新的建模思路.
Abstract:
Idiopathic epiretinal membrane(iERM)is a common vitreoretinal condition caused by the proliferation of glial cells and retinal pigment epithelial cells at the surface of the inner limiting membrane(ILM).Fundus photography,as a low-cost,portable,and easy-to-perform fundus examination,has been widely used for screening fundus diseases.Currently,deep learning models based on fundus photography can identify ERM in affected eyes.However,few studies have quantitatively assessed fundus parameters,particularly retinal vascular morphology,in ERM-affected eyes and use these parameters to predict or assess ERM severity.Optical coherence tomography(OCT)is the gold standard for the clinical diagnosis and assessment of ERM,though due to the high cost and poor portability of the examining devices,it is difficult to carry out large-scale OCT screening in the population.This study aimed to evaluate the correlation between retinal vascular morphological parameters from color fundus photographs and the severity of ERM on OCT scans and explore the viability of using these parameters to predict the severity of ERM in affected eyes.This study retrospectively analyzed retinal vascular morphological characteristics of 292 eyes of 146 iERM patients at Peking University People's Hospital from January 2023 to March 2025 using quantitative technology based on a deep learning-based system EVisionAI.ERM staging based on ectopic inner foveal layer(E1FL),EIFL thickness and central foveal thickness(CFT)on OCT were recorded.Intergroup analyses of vascular morphological parameters between eyes with different stages of ERM were performed and their correlation with ERM staging were analyzed.The results showed:in patients with unilateral ERM,compared with the healthy contralateral eye,the mean vessel branch angle,mean vessel diameter,and mean vessel curvity of the ERM eye were increased,vessel fractal dimension and the angle between the vascular arch and the macula were decreased(P<0.05).The vascular fractal dimension(P=0.004),mean vessel diameter(P<0.001),and mean vessel curvity(P<0.001)were correlated with ERM stages.Vascular fractal dimension less than 1.533(OR=7.247,P<0.001),mean vessel diameter more than 61.93pm(OR=2.277,P=0.013),mean vessel curvity more than 0.893×104(OR=5.044,P<0.001),the angle between the vascular arch and the macula smaller than 88.84°(OR=4.628,P<0.001)were risk factors for stage 3 or 4 ERM.This study analyzed retinal vascular morphological characteristics and their correlation with ERM staging as well as prognostic factors in predicting stage 3 or 4 ERM eyes,providing a new modelling strategy to grade and evaluate the stage and severity of ERM eyes with color fundus images.More prospective studies with larger sample sizes are needed in the future to better explore the relationship between retinal vascular parameters on fundus photographs and the severity as well as surgical prognosis of ERM eyes.
刊 名 科学通报 ISTICEIPKU
年,卷(期): 2026, 71(15)
机标分类号: R774.5 TP393.08 R541.4
在线出版日期: 2026年7月23日
基金项目: 2159000375:星辰黄斑病公益基金会课题,