Macular analysis by multi-contrast segmentation of retinal pigment epithelium and choroid by Jones-matrix OCT

Abstract

Purpose : Current retinal segmentation algorithms are based on layer or boundary delineation. Although it works well with moderately abnormal cases, its applicability to heavily damaged tissue is low. In this study, we demonstrate pixel-wise segmentation method which is applicable equally to normal and pathologic cases.

Methods : A custom made posterior Jones matrix optical coherence tomography is used to obtain multi-contrast images including scattering OCT, attenuation coefficient (AC), birefringence, degree-of-polarization-uniformity (DOPU), and OCT angiography (OCTA). The tissue segmentation is done by applying a threshold to a “feature” which is synthesized from the multi-contrast images as follows. For retinal pigment epithelium (RPE) segmentation, the feature is defined as AC × (1 – DOPU) × (1 –OCTA_b), where OCTA_b is binarized OCTA with Otsu method. The pixel is classified as RPE if the feature is larger than 1.0. For choroidal stromal segmentation, another feature is defined as (1 – DOPU) × OCTA_b. The pixel is classified as choroidal stroma if the feature is larger than 0.1. We also created several en face maps for easy identification of pathology.

Results : 13 eyes of cases were involved. The cases include 4 normal, 4 myopic choroidal neovascularization (mCNV), 2 pigment epithelial detachment (PED), 2 geographic atrophy (GA), and 1 AMD. Fig. 1 shows B-scans obtained from a normal and an mCNV complicated with non-sunset Vogt-Koyanagi-Harada disease (VKH-mCNV) subjects. The first to third columns are AC, DOPU, and OCTA_b. The last column shows segmentation results where RPE and choroidal stroma are displayed as red and green, respectively. The absence of RPE is clearly visible in VKH-mCNV case [Fig. 1(h)]. Fig. 2 shows melano-layer thickness (MLT) maps (a), RPE elevation (RPEE) maps (b), and choroidal thickness (CT) maps (c). In the VKH-mCNV and GA cases, the absence of melano-layer at the center is clearly visible in the MLT map [Fig. 2(a)]. PED is clearly shown in the RPEE maps [Figs. 2(b)]. The CT maps [Figs. 2(c)] show VKH-mCNV subject has thinner choroid than the normal subject.

Conclusions : Multi-functional OCT based segmentation method is developed. It was applicable both to normal and pathologic cases. En face maps based on the segmentation clearly visualized signatures of pathologies.

This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.

 

Fig. 1 Multi-contrast images and segmented images.

 

Fig. 2 Generated en face maps.