Optimizing the Repeatability of Choriocapillaris Flow Deficit Measurement From Optical Coherence Tomography Angiography

To evaluate the impact of processing technique and slab selection on the repeatability of choriocapillaris (CC) flow deficit (FD) measurements as assessed using optical coherence tomography angiography (OCTA) Prospective, cross-sectional study. Healthy subjects were imaged with 4 consecutive 3 × 3-mm OCTA using a swept-source OCT (PLEX elite 9000; Carl Zeiss Meditec). OCTA images were generated using the Max projection, and three 10-μm-thick slabs starting 11, 21, and 31 μm posterior to the automatically segmented retinal pigment epithelial band. The resultant images were binarized using the Phansalkar method with a 43.94-μm radius and then the CCFD% was computed. The intraclass correlation coefficient (ICC) and coefficient of variation (CV) were computed for the 4 acquisitions to assess the repeatability of the CCFD%. This entire analysis was repeated after separately modulating several parameters: (1) Sum instead of the Max projection, (2) retinal pigment epithelial fit instead of the retinal pigment epithelial band as the offset reference, (3) 14.65 and 87.88 μm radius values instead of 43.94 μm. Twenty-four healthy eyes (mean age; 36.4 years) were enrolled. The CCFD% in the 11-21-, 21-31-, and 31-41-μm slabs generated by the Max algorithm and the retinal pigment epithelial band showed high repeatability values (ICCs = 0.963, 0.975, and 911; CVs = 0.05, 0.05, and 0.05, respectively). As most of the cases were confounded with the hypointense region when the 11-21-μm slab was used, however, this slab could not be included in the subsequent analyses. Those values in the 21-31- and 31-41-μm slabs were higher than those of the corresponding slabs by the Sum algorithm (ICC = 0.916 and 0.776; CV = 0.15 and 0.19, respectively) or by the retinal pigment epithelial fit (ICC = 0.907 and 0.802; CV = 0.06 and 0.06, respectively). The Phansalkar radius of 43.94 μm had the highest ICC numerically, but this was not statistically significantly greater than for a radius of 14.65 μm (ICC = 0.960 and 0.911, respectively) or a radius of 87.88 μm (ICC = 0.958 and 0.897, respectively). Regardless of which parameter was modulated, the 21-31-μm slab was the most repeatable. In normal eyes, en face CC OCTA images generated using the Max projection and a 10-μm-thick slab offset of 21 μm below the instrument-generated retinal pigment epithelial band yielded the most repeatable CCFD%. These findings have implications for the design of standardized processing algorithms for quantita