Longitudinal analysis of retinal vessel parameters for enhanced disease development insights and improved risk prediction models

Progression of diabetic retinopathy is closely linked to morphological changes observed in retinal vessels. However, longitudinal variations remain unclear. In this study, we analysed 920 cases, focusing primarily on changes in vessel morphology in relation to diabetic retinopathy. Using the QUantitive Analysis of Retinal vessel Topology and SiZe (QUARTZ) model enabled extraction of quantitative data from fundus images, including vessels size, tortuosity and vessel type. Longitudinal image registration model was applied to precisely match vessels across images captured at different time points. These steps enabled computation of mean vessel diameter and identification of changes in matched vessels over time. The data enabled analysis of two groups: progressors and non-progressors. The progressor group included cases in which diabetic retinopathy grade changed from R0M0 to diabetic retinopathy (R0: no retinopathy; M0: no maculopathy). From this analysis, several observations were made. Arteries appeared thicker than veins. Vessels size was observed to change in both groups. Vessels became thinner in non-progressors and thicker in progressors. Vein size changes appeared greater than arterial changes in progressors, although changes were noted in both vessels types. These findings suggest that vessel changes, particularly vessel thickening, may be associated with progression to diabetic retinopathy. They provide a basis for further research into vessels size characteristics, including location in the retina and proximity to the optic disc, with potential implications for early clinical monitoring and risk stratification. In addition, the longitudinal analysis allowed consistent tracking of the same segments across time, reducing variability associated with independent image analysis. Quantitative measurements were standardised across all images to ensure comparability between time points and study groups. The combination of automated vessel extraction and precise matching provides a scalable methodology for analysis of large datasets. Furthermore, separating arteries and veins allowed a more detailed evaluation of vessel-specific behaviour in relation to diabetic retinopathy progression.