A semantic segmentation method to analyze retinal vascular parameters of diabetic nephropathy

Abstract

Introduction: By using spectral domain optical coherence tomography (SD-OCT) to measure retinal blood vessels. The correlation between the changes of retinal vascular structure and the degree of diabetic nephropathy is analyzed with a full-pixel Semantic segmentation method.

Methods: A total of 120 patients with diabetic nephropathy who were treated in the nephrology department of Quzhou People's Hospital from March 2023 to March 2024 were selected and divided into three groups according to the urinary albumin creatinine ratio (UACR). The groups included simple diabetes group (UACR < 30 mg/g), microalbuminuria group (30 mg/g ≤ UACR <300 mg/g) and macroalbuminuria group (UACR ≥300 mg/g). SD-OCT was used to scan the arteries and veins in the superior temporal area B of the retina. The semantic segmentation method built into the SD-eye software was used to automatically identify the morphology and structure of the vessels and calculate the parameters of arteriovenous vessels. The parameters of arteriovenous vessels are as follows: outer diameter of the retinal artery (RAOD); inner diameter of the retinal artery (RALD); arterial wall thickness (AWT); arterial wall to lumen ratio (AWLR); cross sectional area of arterial wall (AWCSA); retinal vein outer diameter (RVOD); retinal vein inner diameter (RVLD); vein wall thickness (VWT); vein wall to lumen ratio (VWLR); cross sectional area of vein wall (VWCSA). Statistical analysis software was used to compare and analyze the parameters of retinal arteriovenous vessels of the three groups.

Results: The study revealed statistically significant differences in RAOD and RALD among the three groups (p < 0.05) with the RAOD and RALD of the macroalbuminuria group and microalbuminuria group being lower than those of the simple diabetes group. Conversely, there were no significant differences in AWT, AWLR and AWCSA among the three groups (p > 0.05). Additionally, the differences in RVOD and RVLD among the three groups were found to be statistically significant (p < 0.05) with the RVOD and RVLD of the simple diabetes group being lower than those of the microalbuminuria group and macroalbuminuria group. No significant differences were observed in VWT and VWL among the groups. Additionally, RVOD and RVLD were weakly associated with UACR (R = 0.247, p = 0.007; R = 0.210, p = 0.021). Full-pixel semantic segmentation method combined with OCT images is a new retinal vascular scanning technology, which can be used as a new method for early diagnosis of diabetic nephropathy. The structural changes of retinal vessels can be used to predict the severity of diabetic nephropathy during the development of diabetic nephropathy.

Keywords: diabetic nephropathy; full pixel semantic segmentation method; optical coherence tomography; urinary protein creatinine ratio; vascular structure.

Figure 1

operation process (A): The green straight line represents the OCT linear scan line. the shaded (B) area is the linear scan area. When the blue arrow marks the artery, the green arrow marks the vein. B: The image shows the corresponding cross-sectional area image of the arterial and venous vessels in Figure A. (C): The interface is displayed after the initial startup of the Huiyan OCT vascular automatic measurement system (D): the corresponding arterial and venous parameters are measured after importing into the Huiyan software OCT vascular automatic measurement system.