Measurement and characterization of retinal vascular morphology parameters based on artificial intelligence automated analysis technology

Objective

To analyze retinal vascular parameters and distribution characteristics in Chinese population via the fully automated quantitative measurement of retinal vascular morphological parameters based on artificial intelligence technology.

Methods

A cross-sectional study was performed.A total of 1 842 patients without fundus diseases who visited Beijing Tongren Hospital from January 2011 to December 2021 were included.Standardized questionnaires, blood draws and ophthalmologic examinations of enrolled subjects were conducted.Color fundus photographs centered on the optic disk of one eye of patients were collected, and a deep learning-based semantic segmentation network ResNet101-Unet was used to construct a vascular segmentation model for fully automated quantitative measurement of retinal vascular parameters.The main measurement indexes included retinal vascular branching angle, vascular fractal dimension, average vascular caliber, and average vascular tortuosity.To compare different retinal parameters between sexes, the correlation between the above parameters and ocular factors such as best corrected visual acuity, intraocular pressure, and axial length, as well as systemic factors such as sex, age, hypertension, diabetes mellitus, and cardiovascular disease was analyzed.This study adhered to the Declaration of Helsinki.The study protocol was approved by the Ethics Committee of Beijing Tongren Hospital, Capital Medical University (No.20001220). Written informed consent was obtained from each subject.

Results

The model established in this study achieved an accuracy over 0.95 for both vascular and optic disk segmentation.The vascular branching angle, vascular fractal dimension, average vascular caliber, and average vascular tortuosity were (51.023±11.623)°, 1.573(1.542, 1.592), 64.124(60.814, 69.053)μm, (0.001 062±0.000 165)°, respectively.Compared with females, males had larger vascular branching angle, smaller average vascular caliber and smaller vascular tortuosity, and the differences were statistically significant (all at P<0.05). The average vascular caliber increased by 1.142 μm in people with cardiovascular disease compared to people without cardiovascular disease (B=1.142, P=0.029, 95%CI: 0.116-2.167). The average vascular tortuosity was positively correlated with hypertension (B=3.053×10^-5, P=0.002, 95%CI: 1.167×10^-5-4.934×10^-5) and alcohol consumption (B=1.036×10^-5, P=0.014, 95%CI: 0.211×10^-5-1.860×10^-5) and negatively correlated with hyperlipidemia (B=-2.422×10^-5, P=0.015, 95%CI: -4.382×10^-5-0.462×10^-5). For each 1-mm increase in axial length, there was a decrease of 0.004 in vessel fractal dimension (B=-0.004, P<0.001, 95%CI: -0.006–0.002), a decrease of 0.266 μm in the average vessel caliber (B=-0.266, P=0.037, 95%CI: -0.516–0.016), and a decrease of -2.45×10^-5° in the average vessel tortuosity (B=-2.45×10^-5, P<0.001, 95%CI: -0.313×10^-5–0.177×10^-5). For each 1.0 increase in BCVA, there was an increase of 3.992° in the vascular branch angle (B=3.992, P=0.004, 95%CI: 1.283-6.702), an increase of 0.090 in vascular fractal dimension (B=0.090, P<0.001, 95%CI: 0.078-0.102) and a decrease of 14.813 μm in the average vascular diameter (B=-14.813, P<0.001, 95%CI: -16.474–13.153).

Conclusions

A model for retinal vascular segmentation is successfully constructed.Retinal vessel parameters are associated with sex, age, systemic diseases, and ocular factors.