Abstract
Purpose :
Being able to precisely track bloodflow dynamic inv vivo is important in understanding a number of disease of the eye, CNS and systemically. Direct imaging of medium to large vessels is complicated due to relatively slow frame rates and aliasing artifacts, hence the development of XT (X direction by Time) imaging, whereby the slow scanner is paused for a period of the scan creating a normal rate structural image and a fast scan rate over the vessel of interest. The aim of this study is to design an algorithm to extract such measurements automatically from these images.
Methods :
In the temporal portion of XT, blood cells move past the fast scanner resulting in a streak. Depending on the speed of the bood cell, the angle and length of the streak will vary. Streak orientation was selected by sliding a diagonal convolution kernel (consisting of a line of a certain slope) along the XT Scan, the fit of such kernel was evaluated at each pixel location. The speed of erythrocytes was automatically estimated by testing kernels with several slopes and selecting the best fit.
Results :
The angles of lines belonging to the best fitting kernels were plotted against angles estimated by human graders. We evaluated our algorithm on 100 consecutive XT Scan frames from a BMC Apaeros AOSLO, revealing the known fluctuation of blood velocity and flow during the cardiac cycle. Plotting the graders’ estimates against those produced by our algorithm resulted in a Pearson correlation coefficient = 0.70 (P=2E-35).
Conclusions :
Here we present a pipeline to automatically detect time portions of XT images and extract streak angle, thus determining blood flow. Our method compared favorably with human graders, while taking a fraction of the time to compute. This pipeline can be used automatically within further analysis pipleines to determine new insights into diabetes and other vascular diseases.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.