AUTHOR=Guo Yiqin , Sun Yunxiao , Zhang Xueyuan , Wang Ningli TITLE=Performance Assessment of Two Different Approaches of Measuring Skeletonized Radial Peripapillary Capillary Vessel Density in Glaucoma Patients JOURNAL=Frontiers in Medicine VOLUME=8 YEAR=2022 URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2021.814306 DOI=10.3389/fmed.2021.814306 ISSN=2296-858X ABSTRACT=Objective

To compare performance assessment of two methods of measuring radial peripapillary capillary (RPC) vessel density (VD) after skeletonization using MATLAB and Image J in glaucoma clinical setting.

Methods

Seventy-three eyes of 73 glaucoma patients from Beijing Tongren Hospital were included in this prospective study. Original images of RPC were obtained using optical coherence tomography angiography. Two approaches were executed before measuring. Method 1 (M1) required image sharpening, removal of big vessels, and skeletonization. Method 2 (M2) required skeletonization and removal of major vessels. Each method was executed twice. Repeatability and correlations with glaucomatous parameters were assessed. Factors associated with retinal nerve fiber layer thickness (RNFLT) and visual field mean deviation (MD) were analyzed.

Results

Average VD was 13.86 ± 2.73 and 7.50 ± 2.50% measured by M1 and M2. Percentage of total elimination of the major vessels was 36.99 and 100% by M1 and M2, respectively. The intrasession and intersession reproducibility was higher by M2 (ICC = 0.979, ICC = 0.990) than by M1 (ICC = 0.930, ICC = 0.934). VD measured by M2 showed stronger correlations with glaucomatous parameters than by M1. By stepwise multiple linear regression, thinner RNFLT was associated with smaller VD measured by M2 (B = 4.643, P < 0.001). Worse MD was associated with smaller VD measured by M1 (B = 1.079, P = 0.015).

Conclusion

The VD measured by M2 showed better reproducibility and higher correlation with glaucomatous structural parameters. Image sharpning helps display of hazy vasculature in glaucoma, which may reflect visual function better. Researchers should carefully choose image processing methods according to their research object.