Abstract
Purpose :
This study is to establish quantitative features of vascular geometry in optical coherence tomography angiography (OCTA) for objective classification of diabetic retinopathy (DR).
Methods :
OCTA images of 20 control subjects and 60 non-proliferative DR (NPDR) (mild, moderate and severe stages) patients were analyzed for this study. For geometric analysis, individual vessel branches were segmented, and branch angles and calibers were measured. Six geometric features, including total branching angle (θ), branching angle of the larger child vessel segment (α1), branching angle of the smaller child vessel segment (α2), branching coefficient (BC), the larger child to parent vessel width ratio (WR1), and the smaller child to parent vessel width ratio (WR2), were derived from each vessel branch. Figure 1 illustrates these six parameters. In pre-processing steps, a Frangi 2D filter was applied on the OCTA images to enhance the vessels, and global thresholding with morphological functions was used to generate binary vessel and skeleton map. From the skeleton map, each branch point and triangles were automatically segmented. For each of the branch triangles, corresponding endpoints, larger and smaller vessel segments were identified, and the geometric features were calculated. To test the performance of the geometric features, one-way multi-label analysis of variance (ANOVA) was used to compare difference of mean values of the geometric features among control and DR (mild, moderate and severe NPDR) groups. One versus one comparisons of the geometric features between the control and NPDR stages were performed using Student’s t-test.
Results :
Among six geometric features of retinal vasculature in OCTA, total branching angle (θ) and width ratios (WR1 and WR2) were most sensitive to DR stages. Five parameters, θ, α2, BC, WR1 and WR2 had positive correlation with severity of NPDR staging. The α1 parameter did not show any change with DR progression. In case of control vs. NPDR, θ, α2, WR1 and WR2 increased 0.95%, 3.2%, 9.53% and 12.42%, respectively (Student’s t-test: P<0.05 for θ, α2 and WR1; P<0.001 for WR2). WR2 was the only feature that was able to differentiate control and individual NPDR groups (ANOVA, P<0.05).
Conclusions :
Quantitative geometric analysis revealed vascular distortions in OCTA of NPDR patients, promising objective biomarkers for automated classification and treatment evaluation of DR.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.