Improved 3D Modeling of choroidal Haller’s sublayer vasculature based on swept-source OCT scans using Phansalkar thresholding

Abstract

Purpose : Choroid, a dense vascular layer, located posterior to the outer retina, maintains the health of the retina and other posterior segment structures by supplying them with oxygen and nutrients. Structural changes in the choroid are associated with various vision-threatening diseases including age-related macular degeneration (AMD) and central serous chorioretinopathy (CSCR). In particular, clinicians hypothesize that accurate quantification of choroid vessel biomarkers in 3D using optical coherence tomography (OCT) imaging may enable early diagnosis. However, various attempts to detect choroidal vessels in OCT B-scans are suboptimal. Against this backdrop, we attempt to accurately detect choroidal vasculature in high-resolution swept-source OCT (SS-OCT) volumes based on Phansalkar thresholding

Methods : This retrospective study was performed on 15 (5 healthy, 5 AMD, and 5 CSCR) SS-OCT volumes taken from the Carl Zeiss Plex Elite 9000 device. First, we segmented the choroid layer based on our previously validated residual encoder-decoder (ResUNet) deep learning model and 3D smoothing. Subsequently, we binarized each B-scan using phansalkar thresholding (block size: 40), extracted the binarized choroid layer using detected boundaries, stacked binarized choroid from all B-scans and performed 3D open-close to produce smooth 3D vasculature. For comparative analysis, we repeated the steps for our previously reported exponentiation enhancement (EE) based choroid binarization. The accuracy of the vessel segmentation was validated using subjective grading performed on 2D (three randomly chosen B-scans/volume) and 3D vasculature of Phansalkar and EE methods.

Results : The mean 2D and 3D grading scores for the Phansalkar-based vasculature are observed to be 92.67% and 94% while for the EE method, they are observed to be 77.67% and 73.33%, respectively. 2D and 3D choroid vasculatures by the two methods indicate that the boundary and density of choroidal vessels are more accurate in the proposed method (Figures 1 and 2).

Conclusions : The proposed Phansalkar method significantly improves the performance of choroidal vessel segmentation.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

Figure 1: Phansalkar vs EE: 2D choroidal vasculature.

 

Figure 2: Phansalkar vs EE: 3D choroidal vasculature.