Quantification of choroidal Haller’s sublayer vasculature in 3D based on wide-field SS-OCT scans using 3D tensor voting and geometric modeling

Abstract

Purpose : The choroid is the dense vascular layer present posterior to the outer retina, serving various metabolic functions such as supplying oxygen and nutrients to the retina layers. Studies indicate that choroidal structural changes are related to several vision-threatening posterior segment disorders. Clinicians hypothesize that early manifestations of retinal diseases may be localized to specific regions of choroidal vasculature. However, studies so far were based on gross choroidal biomarkers including the thickness, volume, and vascularity index but not based on finer vasculature quantification. In this regard, to enable early diagnosis, clinicians seek to have biomarkers at the level of blood vessels in 3D. In response, we proposed and validated an end-to-end methodology to quantify choroidal Haller’s sublayer vasculature in 3D using wide-field SS-OCT volumes.

Methods : This is a retrospective study performed using healthy and diseased wide-field SS-OCT volume scans taken from the Carl Zeiss Plex Elite 9000 device. We proposed an end-to-end algorithm to quantify Haller’s sublayer vasculature. Firstly, we employ previously validated 3D residual U-Net, binarization based on exponential and non-linear enhancement, and 3D smoothing to segment the vasculature. Secondly, we employ TEASAR, 3D tensor voting, and geometrical modeling to estimate centerline (CL) and cross-sectional (CS) radius (Figure 1(a)). The accuracy of CL and CS estimates was validated based on subjective grading, performed on both synthetic and choroidal (of one healthy and two diseased eyes) vasculature. In each session (total 3 sessions), the grader graded CS at 5 randomly picked CL points (Figure 1(b)) to measure various parameters (Figure 1(c)).

Results : For each of the four parameters under consideration, we achieved an average grading score above 92%, demonstrating the efficacy of the proposed methodology. 3D heatmaps depicting relative CS radius as well as histograms of the representative healthy and diseased eyes facilitate quantitative visualization of the vasculature (Figure 2).

Conclusions : The proposed methodology is accurately quantifying choroidal Haller’s sublayer vasculature in 3D.

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

 

(a) Proposed methodology; (b) Subjective grading; and (c) Subjective grading scores on synthetic vessel models.

 

CS radius quantification represented by heatmap for healthy and diseased eyes.