人工智能辅助下家族性渗出性玻璃体视网膜病变的血管形态研究

Vascular Morphology Study of Familial Exudative Vitreoretinopathy Assisted by Artificial Intelligence

摘要: 目的：探索家族性渗出性玻璃体视网膜病变早期筛查及诊断的量化指标。方法：采用回顾性病例对照设计，纳入FEVR患者34例，对照组32例，收集超广角眼底照相，实现血管网络自动化分割，计算分形维数(fractal dimension, FD)，对比组间血管拓扑复杂度差异。结果：建立了FEVR的人工智能辅助诊断框架，显著提升了对FEVR周边血管病变的识别效能，发现在超广角眼底图像中FEVR患者及正常对照者的视网膜血管的FD值存在差异(P < 0.05)。结论：VGG16 U-Net模型应用于FEVR患者眼底血管分形特征的研究有望在更广泛的临床应用中发挥作用，提示分形维数可作为FEVR疑诊患者早期筛查及诊断的量化指标。

Abstract: Objective: To explore the quantitative indicators of early screening and diagnosis of familial exudative vitreoretinopathy. Methods: A retrospective case-control design was adopted to include 34 FEVR patients and 32 control patients. Ultra-wide-angle fundus photography was collected to realize automated vascular network segmentation, fractal dimension (FD) was calculated, and the difference of vascular topology complexity between groups was compared. Results: An AI-assisted diagnosis framework for FEVR was established, which significantly improved the recognition efficiency of peripheral vascular lesions in FEVR. There were differences in FD values of retinal blood vessels between FEVR patients and normal controls in ultra-wide-angle fundus images (P < 0.05). Conclusion: The application of VGG16 U-Net model in the study of fractal features of fundus blood vessels in FEVR patients is expected to play a role in a wider range of clinical applications, indicating that fractal dimension can be used as a quantitative indicator for early screening and diagnosis of FEVR patients in doubt.