Abstract
Purpose :
Vessel density (VD) from optical coherence tomography angiography (OCTA) images is a potentially important clinical endpoint in retinal diseases but is dependent on the thresholding method utilized. To evaluate this effect, we compare several thresholding methods in a cohort of normal subjects.
Methods :
In this prospective IRB approved study, 17 healthy subjects underwent OCTA imaging. 12x12 mm macular scans were obtained with swept-source OCTA (PlexElite 9000, Zeiss, Dublin, CA). Scans with signal strength <8 and >4 motion artifacts were excluded. VD was calculated for each scan using three global thresholding methods (Default, Mean, Otsu) and one local method (Phansalkar, radius of 1/32 of total pixel diameter) in ImageJ (NIH, Bethesda, MD) and compared against the Advanced Retina Imaging (ARI) Network algorithm (Zeiss). ANOVA and Tukey’s range tests comparing the calculations of each automated thresholding algorithm were performed.
Results :
17 eyes were included in analysis. The mean VD was 39.17 ± 4.88 mm2 by the default method, 44.64 ± 2.17 mm2 by mean, 38.89 ± 4.18 mm2 by Otsu, 48.09 ± 3.85 mm2 by Phansalkar, and 33.49 ± 4.41 mm2 by ARI. ANOVA showed a statistically significant difference among methods (p < 0.0001). The pairwise comparison yielded statistically significant differences (p < 0.0001) between all groups, except for default vs. Otsu.
Conclusions :
This is the first study comparing thresholding algorithms in wide field OCTA images. Our findings demonstrate the influence of thresholding methods in the quantitative assessment of vessel density from OCTA images. These results highlight the importance of utilizing the same algorithm for longitudinal comparisons.
This is a 2021 Imaging in the Eye Conference abstract.