Automated Case-Adaptive Slab Configuration for Visualization of Choroidal Neovascularization in Swept-Source OCT Angiography

Abstract

Purpose : The extent of choroidal neovascularization (CNV) is difficult to assess and quantify in swept-source OCT-angiography (SSOCT-A) due to the high signal from the surrounding choriocapillaris (CC). We propose a fully-automated method to isolate the CNV signal and produce an angiography slab to visualize it easily.

Methods : Retinal volumes of 3x3mm and 6x6mm field of view were collected using a PLEX™ Elite 9000 with AngioPlex™ OCT Angiography (ZEISS, Dublin, CA). A pre-processing step corrected decorrelation tail artifacts (also known as projection artifacts) throughout the entire SSOCT-A volume. A volumetric segmentation of the CNV, isolated from the CC and other vessels, is generated automatically by compressing an initial estimate of the outer retina into regions that most resemble the vessel-like structure expected in CNV. The volumetric CNV and CC signals are combined in a color-coded en face slab together with the superficial vessels. An additional slab is also generated where colors indicate different CNV retinal depth. The resulting images were compared to traditional slabs (using segmentation derived from structure) of the same eyes.

Results : We processed 28 SSOCT-A scans from eyes with CNV (including scans acquired longitudinally through time). For comparison, we also collected 10 scans from eyes with dry age-related macular degeneration (AMD) and 27 scans from healthy eyes. Our method correctly identified a volumetric region in all of the CNV scans while correctly not highlighting any region in dry AMD or healthy scans. All images generated by our method showed improved visualization of CNV compared to traditional slabs, especially in cases with high CC signal surrounding the CNV. While the extent of the CNV could be difficult to judge in the traditional slabs, the images generated by our method showed distinct CNV outlines, with the added possibility of displaying CNV depth using different colors (see Figure 1).

Conclusions : Our method generates color-coded en face images where CNV can be easily visualized, outlined, quantified, and followed over time.

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

 

Figure 1: Resulting images for two different neovascular lesions. Left: Color-coded retinal vessels, with CNV vessels in green. Center: CNV color-coded by depth (green and red) within the CC (blue). Right: B-scans indicated by the red lines in the en face images, with superimposed CNV depth angiography signal.