Visualization of Choroidal Neovascularization in Optical Coherence Tomography Angiography using Automatic Segmentation and Manual Editing Propagation Tools

Abstract

Purpose : Optical Coherence Tomography Angiography (OCT-A) facilitates the visualization of choroidal neovascularization (CNV). Accurate segmentation of the outer plexiform layer (OPL) and Bruch’s membrane (BRM) is essential for the enface visualization of the CNV net. However, the detection of the retinal layers in pathological eyes is difficult. We aim to investigate the capability of automatic segmentation and manual editing propagation tools of AngioVue (Optovue, Inc., Fremont, CA, USA) for the visualization of CNV.

Methods : In this study, we evaluated nine OCT-A scans of 3x3mm² field size (304x304 A-scans) centered on the fovea from 9 patients with CNV due to Age Related Macular Degeneration. Each scan was processed with the built-in automatic segmentation tool and 8 retinal surfaces were generated in 10 secs. Projection artifact removal (PAR) was applied to the 3D volume and the resulting enface image based OPL to the BRM slab was used for CNV visualization. Two graders reviewed the results of OPL and BRM independently and made corrections where needed with manual editing propagation tool.

Results : The OPL was correctly detected without editing in all 9 scans and BRM was correctly detected without editing in 4 scans and CNV could be visualized clearly as illustrated in Fig. 1. In the other 5 cases, the automatically detected BRM curve was misplaced and CNV visualization was partial in the enface slab (Fig. 2a). After manual editing of BRM (typically modifying 3~5 B-scans) and propagating through the entire volume (usually < 1 sec), the entire CNV net could be visualized by both observers (Fig. 2b and 2c). The inter-observer differences of average OPL-BRM thickness using manual editing propagation tool in 5 scans were 3.1 ± 3.1 µm.

Conclusions : The combination of automatic segmentation and manual editing propagation tool can assist in reliable detection of OPL and BRM surface in CNV subjects, facilitating CNV identification and possibly quantification in the future.

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.

 

OCT B-scan through CNV (a) same scan overlaid with OCTA signals and OPL and BRM curve (b) with corresponding enface slab (d). Same b-scan with OPL-BRM curves following PAR (c) and corresponding OCT-A enface slab (e).

 

The enface images of OPL-BRM slab and corresponding B-Scans after PAR using automatic segmentation, grader 1 edits and grader 2 edits of BRM (a,b and c, respectively).