Assessing the Effect of Image Scaling on OCTA Vessel Density Deviation Mapping

Abstract

Purpose : Differences in axial length have been shown to alter foveal avascular zone area measures on OCTA images.^1,2 We assessed the effect of differences in axial length on OCTA-based deviation mapping of vessel density.

Methods : Subjects with no known ocular or systemic pathology ("control", n = 340), macular telangectasia ("MacTel", n = 18), diabetes without retinopathy (n = 14), nonproliferative diabetic retinopathy (n = 19), and proliferative diabetic retinopathy (n = 15) underwent OCTA imaging and axial length measurement. A single nominal 3x3mm vasculature slab (inner limiting membrane to 9 µm below the outer plexiform layer) and a pixel-by-pixel density map for each subject were used. Two normative databases were constructed using density maps from control subjects, one using the device's model eye length (23.95mm) and the other using subjects’ axial lengths. For each patient with pathology, their density maps were analyzed using each of the normative databases (calculating the percentage of their density maps that were more than two standard deviations from the mean of the database).

Results : Mean (± SD) axial length was 24.23 ± 1.24 mm for control subjects, 22.99 ± 0.94 mm for patients with MacTel, and 23.55 ± 0.92 mm for all patients with diabetes. The use of scaled reference data lead to a greater absolute mean total deviation compared to unscaled reference data (0.24 ± 1.00%, Wilcoxon signed-rank test, p = 0.026). Relative differences between the use of scaled versus unscaled reference data were much greater (Figure); patients with MacTel had a relative difference of 4.1 ± 10.6% (range: -13.41 to 21.39%), while those with diabetes had a relative difference of 0.12 ± 6.98% (range: -29.17 to 17.95%).

Conclusions : Unscaled OCTA image databases result in significant underestimation of total vessel density deviation. In subjects with low (but measureable) overall deviation, differences reached 30% between the scaled and unscaled deviation maps. Correct scaling has greater impact in detecting early vessel damage or small changes in vessel density, though larger overall deviations can be detected using either scaled or unscaled data.

¹PMID: 28616362
²PMID: 28622398

This is a 2020 ARVO Annual Meeting abstract.

 

Relative percent difference of the total deviation area versus the average deviation area. Bias (solid line) was 1.20% with 95% CI of 0.99-1.41% (shaded area). Data plotted separately for subjects with MacTel (crosses) and those with diabetes (open squares).