Comparative analysis of retinal vascular structural parameters in populations with different glucose metabolism status based on color fundus photography and artificial intelligence.

Abstract

Objective: Measure and analyze retinal vascular parameters in individuals with varying glucose metabolism, explore preclinical retinal microstructure changes related to diabetic retinopathy (DR), and assess glucose metabolism's impact on retinal structure.

Methods: The study employed a cross-sectional design encompassing a 4-year period from 2020 to 2024. Fundus photographs from 320 individuals (2020-2024) were categorized into non-diabetes, pre-diabetes, type 2 diabetes mellitus (T2DM) without DR, and T2DM with mild-to-moderate non-proliferative DR (NPDR) groups. An artificial intelligence (AI)-based automatic measurement system was used to quantify retinal blood vessels in the fundus color photographic images, enabling inter-group parameter comparison and analysis of significant differences.

Results: Between January 2020 and June 2024, fundus color photographs were collected from 320 individuals and categorized into four groups: non-diabetes (n = 54), pre-diabetes (n = 71), T2DM without overt DR (n = 144), and T2DM with mild-to-moderate NPDR (n = 51). In pairwise comparisons among individuals with pre-diabetes, T2DM without DR, and T2DM with mild-to-moderate NPDR. Fasting blood glucose (FBG), glycated hemoglobin (HbA1c), systolic blood pressure (SBP), and diastolic blood pressure (DBP) were significantly different (P < 0.05). Within the T2DM population, FBG, HbA1c, age, SBP, and DBP were significant predictors for mild-to-moderate NPDR (P < 0.05). Average venous branching number (branch_avg_v) was significantly different among pre-diabetes, T2DM without DR, and T2DM with mild-to-moderate NPDR groups. In patients with T2DM with mild-to-moderate NPDR, Average length of arteries (length_avg_a) and average length of veins (length_avg_v) increased, whereas branch_avg_v, average venous branching angle (angle_avg_v), average venous branching asymmetry (asymmetry_avg_v),overall length density (vessel_length_density), and vessel area density (vessel_density) decreased significantly (P < 0.05). Logistic regression analysis identified length_avg_a, branch_avg_v, angle_avg_v, asymmetry_avg_v, vessel_length_density, and vessel_density as independent predictors of mild-to-moderate NPDR in patients with T2DM. Receiver Operating Characteristic (ROC) curve analysis demonstrated that length_avg_a, length_avg_v, branch_avg_v, angle_avg_v, asymmetry_avg_v, vessel_length_density, and vessel_density had diagnostic value for mild-to-moderate NPDR (P < 0.05).

Conclusion: In individuals diagnosed with T2DM, specific retinal vascular parameters, such as branch_avg_v and vessel_density, demonstrate a significant correlation with mild-to-moderate NPDR. These parameters hold promise as preclinical biomarkers for detecting vascular abnormalities associated with DR.