Abstract
Purpose :
We examine the ability of Minkowski functional based morphometrics to quantify the 3D structure of retinal blood vessels. We compare data from the same patch of retina that was sampled by OCT angiography (OCTa), and histology.
Methods :
A freshly enucleated porcine eye was cannulated at a cilio-retinal artery supplying the superior hemisphere of the porcine retina and perfused with red blood cells. OCTa signals were collected with an updated version of a custom-built OCT prototype and data was collected from the single porcine retina during perfusion. After collecting the OCTa data the eye was perfused and labelled intravascularly using Lectin conjugated to FITC and the retina was flat mounted for optical sectioning using confocal microscopy (Yu, P.K., et al., 2010. IOVS 51, 447-458). Both data sets were segmented into 3D tubular models. The histological data were higher resolution so were parsed into 3x6 segments, and the OCTa data into 2x4 segments. Each segment had 38 depth planes. The 2D Minkowski functionals (MFs): the area, perimeter and the 4-way and 8-way Euler number were calculated for each depth of each segment. The MFs quantify progressively higher-order correlations between pixels, up to 4th order for the Euler numbers. We then entered the 4 MFs per depth and segment into a principal component based factor analysis.
Results :
Four-factor models captured 0.95 ± 0.06 of the variance in the 38 layers of sub-region MF data of the OCTa data, and 0.86 ± 0.09 of the microscope data (median ± SD). The median proportion of variance accounted for at each depth was 95.3 ± 6.2 %. The factor loadings on depth were remarkably similar for the two data types (Fig. shows factors 1 to 3), with correlation between the loadings of r=0.99, 0.94, 0.91, and 0.93 (all p < 10-12). The factors picked up the 3 layers of the retina, retinal depth, and the borders between layers. Factor 1 was a contrast between the Euler numbers and the area and perimeter, and so contained only 3rd and 4th order correlations between neighbouring pixels. All factors were well weighted onto all MFs.
Conclusions :
The fact that remarkably similar factors were extracted despite very different segmentation of the data illustrates the utility of the additive nature of the MFs. Apparently similar information about retinal vessel structure can be obtained from histological and OCTa data.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.