Abstract
BACKGROUND: Gestational diabetes mellitus (GDM) is a temporary metabolic disorder in which small retinal vessels may have experience subtle changes before clinical lesions of the fundus retina appear. An innovative artificial intelligence image processing technology was applied to locate and analyze the small retinal vessel morphology and accurately evaluate the changes of the small retinal vessels in GDM patients and pregnant women with normal blood glucose and non-pregnant women with normal blood glucose. METHODS: The subjects were divided into three groups:GDM group, pregnant control group (PC), and normal control group (NC). Use optical coherence tomography angiography (OCTA) to collect OCT images of subjects,and perform quantitative identification and analysis of retinal vessel parameters based on artificial intelligence measurement software integrated the prior knowledge supervised edge-aware multi-task network (PKSEA-Net): Retinal arteriolar lumen diameter (RALD), retinal arteriolar outer diameter (RAOD), retinal venular lumen diameter (RVLD),retinal venular outer diameter (RVOD),arterial wall thickness (AWT),venular wall thickness (VWT),arterial wall to lumen ratio (AWLR),venular wall to lumen ratio (VWLR),arterial wall cross-sectional area (AWCSA),venular wall cross-sectional area (VWCSA), arteriovenous ratio (AVR). RESULTS: This study revealed significant differences in RVOD, RVLD, VWT, VWCSA and AVR between the GDM group and the PC group (p = 0.005, p < 0.027, p = 0.008, p = 0.001, p = 0.022), significant differences in RVOD, RVLD, VWT, VWCSA and AVR between the GDM group and the NC group (p < 0.001, p = 0.001, p < 0.001, p < 0.001, p = 0.001). In GDM group, RVOD, RVLD, VWT and VWCSA increased, while AVR decreased. There were no significant differences in RVOD, RVLD, VWT, VWCSA and AVR between PC group and NC group (p = 0.139, p = 0.263, p = 0.107, p = 0.059, p = 0.218), and no significant differences in VWLR among the three groups (p > 0.05). No significant difference was observed in retinal artery vascular parameters (RAOD, RALD, AWT, AWLR, AWCSA) across the three groups (p > 0.05). CONCLUSION: There were increases in RVOD, RVLD, VWT, and VWCSA, decrease in AVR in patients with GDM. However, no significant difference of retinal vascular parameters was shown between normal pregnant women and normal non-pregnant women. PKSEA-Net can assist to identify changes in retinal vascular morphology and diagnose micro-vascular lesion early in normal pregnant women and high-risk groups of GDM.