Skeletonization reduces the reliability of OCTA retinal blood flow analyses.

Abstract

Purpose : Variable methods of analysing optical coherence tomography angiography (OCTA) images makes comparing different OCTA studies challenging. We performed a prospective observational study investigating agreement between retinal blood flow metrics calculated using different techniques and within different analysis packages. We hypothesised that measures assessing an underlying physiological parameter would show high agreement, whilst measures that did not relate to the patients’ underlying physiology would not.

Methods : We recruited patients undergoing major upper gastrointestinal surgery with planned post-operative care in the intensive therapy unit (ITU) and individuals already admitted to the ITU with sepsis. OCTA scans from 31 right eyes (24 males, 7 females) and 29 left eyes (24 males, 5 females) were included in the final analysis. Custom metrics from MatLab (vessel density [VD], fractal dimension [FD], and their skeletonised analysis [SVD and SFD respectively]) and Heidelberg Engineering ([HE]; Mean vessel length density and Skeletonised analysis) were compared for agreement using intraclass correlation coefficient (ICC).

Results : There was strong agreement between HE Mean and Matlab VD measures, reaching an ICC of 0.84, and moderate agreement between HE Skeleton and Matlab VD with ICCs between 0.48-0.65 depending on eye and vascular layer. There was no agreement between any HE measures and FD with ICCs between 0.24-0.49, or between any of the metrics and Matlab skeletonised values with ICCs as low as 0.04. There was variable agreement between eyes, ranging from 0.28 for SFD to 0.90 for VD.

Conclusions : Our results showed moderate to good agreement between the metrics of the different analysis packages, except for FD, and that skeletonising OCTA scans for retinal blood flow analysis reduced agreement between analysis packages and techniques. The low agreement between HE and FD suggests that these measures assess different and unrelated blood flow parameters. The poor agreement from the skeletonised values suggest that non-skeletonised metrics may be preferred and reported in OCTA studies.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.