Early Diabetic Retinopathy Evaluation With OCTA: A Study on Vascular Branching and Fragmentation

Abstract

Purpose: The purpose of this study was to quantitatively evaluate the branching patterns and vascular fragmentation features in preclinical and early diabetic retinopathy (DR) using optical coherence tomography angiography (OCTA).

Methods: OCTA metrics, including branch node number (BNN), branch node density (BND), end point number (EPN), end point density (EPD), fragmented vessel segment count (FVSC), and fragmented vascular length ratio (FVLR), were measured in foveal and parafoveal regions within superficial and deep vascular plexus (SVP and DVP) in the retina.

Results: Compared to healthy control (HCs), both BNN and BND exhibited a significant decrease in individuals with mild DR across both retinal layers, and also in diabetes mellitus without DR (no DR) within DVP. EPD showed a significant increase in mild DR cases compared to HCs, except for the foveal region in SVP; however, EPN did not demonstrate a significant difference among the three groups. Increases in both FVSC and FVLR were significant across all areas and in both layers of the retina. Notably, these metrics showed more pronounced differentiation in the DVP than the SVP.

Conclusions: Foveal BND and BNN in DVP reveal vascular alterations indicative of preclinical DR. Indicators such as EPD, FVSC, and FVLR in DVP correlate with early DR changes and are useful for its early detection. These initial findings demonstrate the potential and benefits of these quantitative OCTA indices for delineating DR-associated alterations in the retinal microvasculature, indicating their potential clinical utility for improved DR screening.

Figure 1.

Overview of data processing procedures of SVP and DVP. Original OCTA image (A and G), binarized vascular image with FAZ boundary, 1.5 mm, and 2.5 mm circles highlighted in red (B and H), Vascular skeleton image (C and I), VD image (D and J), vascular skeleton image with vessel branch terminals labeled and enlarged view of the red box in E and K (F and L). In Figures E, F, K, and L, orange dots represent branch nodes and blue dots represent endpoints.

Figure 2.

Representative OCTA images in a 3 × 3-mm area around the fovea. The first row: OCTA images; The second row: binarized images; The third row: VD images; The Last row: vascular skeleton images.

Figure 3.

Representative processed OCTA images revealing branch nodes and endpoints in HC, noDR, and mild DR groups in DVP. The first row shows the original branch node- and endpoint-labeled images of three groups. The second and the third rows show the enlarged images of regions 1 and 2, respectively, from the corresponding 3 × 3 mm images. Branch nodes are marked with red dots and endpoints are labelled with blue dots.

Figure 4.

BNN and EPN measured in fovea and parafovea areas in SVP or DVP layer of retinal OCTA images acquired from three groups. BNN (A and B) and EPN (C and D) in the fovea (left Y axis) and parafovea (right Y axis) in SVP (left column) and DVP (right column) were calculated and compared among HC, noDR, and mild DR groups. Data were expressed as mean ± SD; Significant, * < 0.05, ** < 0.01, *** < 0.001. BNN, branch node number; EPN, endpoint number.

Figure 5.

BND and EPD measured in fovea and parafovea areas in SVP or DVP layer of retinal OCTA images acquired from three groups. BND (A and B) and EPD (C and D) in fovea and parafovea in SVP (left column) and DVP (right column) were calculated and compared among HC, noDR, and mild DR groups. Data were expressed as mean ± S.D.; Significant, * < 0.05, ** < 0.01, *** < 0.001. BND, branch node density; EPD, endpoint density.

Figure 6.

FVSC and FVLR measured in fovea and parafovea areas in the DVP layer of retinal OCTA images acquired from three groups. FVSC (A) and FVLR (B) in fovea and parafovea were calculated and compared among HC, noDR, and mild DR groups. Data were expressed as mean ± SD; Significant, * < 0.05, ** < 0.01, *** < 0.001. FVSC, fragmented vessel segment count; FVLR, fragmented vascular length ratio.