Analysis of image features of fundus blood vessel in healthy human eye based on deep learning techniques

Abstract

AIM:To explore the fundus vascular characteristics of healthy individuals based on deep learning techniques, with a view to discovering the range of normal values of the fundus arteries and veins, as well as the relationship between physiological factors, such as gender, age, body mass index(BMI), blood pressure, and fundus vasculature characteristics.METHODS:Fundus images of healthy people were taken from a professional fundus camera, and the subject's blood pressure and laboratory test was collected. Additionally, the fundus arteries and veins were segmented by the improved U-Net model, and the color, morphology and Haralick texture features of the vessels were extracted from computer vision technology.RESULTS:A total of 4 487 cases fundus images were taken and 326 cases with healthy and clear fundus images were screened, including 200 males and 126 females. There were differences in the morphology, color, and textural characteristics of the left and right eyes, as well as of the fundus arterioles and veins, with a mean vessel width(width)of 1.146 in the arteries and 1.430 in the veins, and an arteriovenous ratio about 4:5. Fundus artery and vein characteristics in healthy individuals of different ages(21-30, 31-40, 41-50): compared with the healthy population aged 21-30 and 31-40 years, arterial and venous inverse difference moment(idm), f12 and venous angular second moment(asm)values increased, and arterial and venous contrast(con), entropy(ent), difference entropy(den), and venous sum entropy(sen)values decreased in 41-50 years. Compared with the 21-30 years age group, arterial f12 values increased and venous con values decreased in 31-40 years(all P0.05). Fundus characteristics of healthy people with different degrees of blood pressure: there were statistically significant differences in fundus arteriosus area, width, and venous con, idm, dva, and den values between the normal blood pressure and high blood pressure groups(all P<0.05).CONCLUSION: The characteristics of the left and right eyes as well as the fundus arteries and veins differ in healthy individuals and correlate with physiological factors such as gender, age and blood pressure, which have the value of a potential microcirculation marker.

Keywords

• healthy people
• fundus image
• deep learning
• fundus vasculature(arteries
• veins)