人工智能在糖尿病视网膜病变患者视网膜血管参数分析方面的应用

Application of Artificial Intelligence in the Analysis of Retinal Vascular Parameters in Patients with Diabetic Retinopathy

摘要: 目的：采用人工智能眼底检查系统分析正常人与糖尿病视网膜病变非增殖期(NPDR)患者在视网膜血管参数之间的差异。方法：收集2025年3月1日至2026年3月1日在我院就诊的正常人眼底彩色照片158眼，NPDR患者眼底彩色照片158眼，每只眼均采集以视盘为中心的眼底彩照。所有病例图像均由人工智能眼底检查分析系统对视网膜动脉当量(RAE)、视网膜静脉当量(RVE)、视网膜动静脉比(AVR)等进行标注，观察两组之间视网膜血管参数方面的差异。结果：利用人工智能对正常人组与NPDR组的视网膜血管参数进行分析比较，两组之间RAE相比，NPDR组大于正常人组(t = −19.83, P = 0.000)，两组之间RVE相比，NPDR组大于正常人组(t = −22.15, P = 0.000)，两组之间AVR相比，NPDR组小于正常人组(t = 7.02, P = 0.000)，均有显著性差异，差异有统计学意义。结论：利用人工智能系统进行眼底筛查，可以精确定量测量NPDR患者视网膜血管管径变化，这些参数变化与NPDR存在显著关联，可能作为糖尿病视网膜疾病早期识别的潜在标志物，对实现慢病管理窗口前移、解决医疗卫生资源呈现的地区不平衡问题有重要意义。然而，本研究为横断面设计，因果关系尚需纵向研究进一步验证。

Abstract: Objective: To analyze the differences in retinal vascular parameters between normal subjects and patients with non-proliferative diabetic retinopathy (NPDR) using an artificial intelligence-based fundus examination system. Methods: A total of 158 fundus color photographs from normal subjects and 158 from NPDR patients, collected between March 1, 2025, and March 1, 2026, at our hospital, were included in the study. Each eye underwent fundus photography centered on the optic disc. All images were analyzed by an AI-powered fundus examination system to measure retinal arterial equivalent (RAE), retinal venous equivalent (RVE), and arteriovenous ratio (AVR). Differences in retinal vascular parameters between the two groups were evaluated. Results: Comparative analysis using AI revealed significant differences between the normal and NPDR groups. The RAE was greater in the NPDR group than in the normal group (t = −19.83, P = 0.000). Similarly, the RVE was greater in the NPDR group (t = −22.15, P = 0.000). In contrast, the AVR was smaller in the NPDR group (t = 7.02, P = 0.000). All differences were statistically significant. Conclusion: The use of an AI system for fundus screening enables precise quantitative measurement of changes in retinal vascular diameter in NPDR patients, these parameter changes are significantly associated with NPDR and may serve as potential biomarkers for early recognition of diabetic retinopathy. This approach is significant for advancing the window of chronic disease management and addressing regional imbalances in healthcare resources. However, due to the cross-sectional design of this study, causal relationships need to be further validated by longitudinal studies.