Abstract
Purpose :
Retinal blood velocity can be determined by measurement of erythrocyte streak angles relative to vessel angles on Adaptive Optics Scanning Laser Ophthalmoscopy (AOSLO) space-time or XT images, in which successive one-dimensional scans of retinal vessels are stacked. Variability in manual erythrocyte angle determination may contribute error to blood velocity assessment. We evaluated inter-rater concordance between manual and semi-automated measurements of erythrocyte streak angles on AOSLO XT images.
Methods :
We implemented a novel semi-automated method in which regions of interest are manually drawn around erythrocyte streaks. An algorithm automatically performs Yen thresholding and weighted least squares regression to determine streak angles measured against the horizontal. The three shallowest angles are averaged to enable maximal blood velocity calculations. For comparison, two reviewers manually measured and averaged the three shallowest angles of each XT image using ImageJ software. Bland-Altman plots and concordance correlation coefficients were evaluated to compare measurement methods.
Results :
Our analysis included 16 XT image frames from 4 eyes of 4 patients. The mean difference in average angle measurement was -3.07 and -1.75 degrees for the semi-automated method compared to the first and second reviewers, respectively, indicating that the semi-automated method identified shallower erythrocyte angles than manual reviewers. The mean difference was -0.58 degrees when comparing the semi-automated method with the minimum of the two angles measured between the two manual reviewers. The concordance correlation coefficient was 0.92 for the two manual reviewers, 0.93 and 0.97 for the semi-automated method and the respective manual reviewers, and 0.99 for the semi-automated method and the smaller of the two manual measurements.
Conclusions :
By evaluating a range of streaks in an individual AOSLO XT image, the semi-automated method more consistently identified the shallowest erythrocyte streak angles than manual evaluators. Implementation of this semi-automated workflow may improve measurement consistency, increase throughput, and reveal more insights about retinal blood flow metrics.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.