Differences in Mean Values and Variance in Quantitative Analyses of Foveal OCTA Imaging

PURPOSE: Multiple approaches for quantifying parameters such as vessel density (VD) and vessel length density (VLD) in optical coherence tomography angiography (OCTA) en-face segmentations are currently available. While it is common knowledge that data gathered from different methods should not be directly compared to each other, a comparison of the different methods can help to further the understanding of differences between different methods of measurement. Here we compare a common method of semiautomatically quantifying VD and VLD with an automated method supplied by the manufacturer of an OCTA device and report on differences in performance in order to probe for and highlight differences in values gathered by both methods. METHODS: OCTA was performed using the swept source PLEX Elite 9000 device, software version 2.0.1.47652 (Carl Zeiss Meditec Inc., Dublin, CA, USA). Scans of 3 mm × 3 mm from healthy volunteers centred on the fovea were acquired by a well-trained certified ophthalmologist. Scans with a signal strength of 8 out of 10 or higher were included. Quantitative parameters of the 3 mm × 3 mm cube scans were automatically generated and segmented into superficial capillary plexus (SCP) and deep capillary plexus (DCP) layers using layer segmentation produced by the instrument software and prototype analysis VD quantification software (Macular Density v.0.7.1, ARI Network Hub, Carl Zeiss Meditec Inc., Dublin, CA, USA) supplied by the manufacturer. An alternative approach of quantitative analysis of VD and VLD was performed manually with ImageJ (National Institutes of Health, Bethesda, Maryland, USA), as previously reported. VD was assessed as the ratio of the retinal area occupied by vessels. VDL was measured as the total length of the skeletonised vessels using 1-pixel centre line extraction of the blood vessels. RESULTS: We report differences in standard deviation (SD) in OCTA parameters obtained using different methods. The standard deviation of VD and VLD measurements was statistically significantly different in VD of 3 mm × 3 mm DCP (p = 0.009), VLD of 3 mm × 3 mm SCP (p = 0.000), and VLD of 3 mm × 3 mm DCP (p = 0.021). No statistically significant differences were found in VD of 3 mm × 3 mm SCP (p = 0.128) or VLD of 3 mm × 3 mm SCP (p = 0.107). CONCLUSIONS: As expected, we were able to demonstrate significant differences in quantitative OCTA parameters gathered from the same images using different methods of quantification. Values gathered using different methods are not interchangeable. In scientific studies and in situations where long-term follow-up is necessary, the same device and the same method of quantification should be used to maintain retrospective comparability of measurements