July 2019
Volume 60, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2019
Automated volumetric choroidal neovascularization segmentation and quantification in swept-source OCT angiography using machine learning
Author Affiliations & Notes
  • Luis De Sisternes
    Carl Zeiss Meditec Inc., Dublin, California, United States
  • Homayoun Bagherinia
    Carl Zeiss Meditec Inc., Dublin, California, United States
  • Katherine Makedonsky
    Carl Zeiss Meditec Inc., Dublin, California, United States
  • Ala El Ameen
    Créteil University Hospital, France
  • Giovanni Gregori
    Ophthalmology, Bascom Palmer Eye Institute, Miami, Florida, United States
  • Philip Rosenfeld
    Ophthalmology, Bascom Palmer Eye Institute, Miami, Florida, United States
  • Mary K Durbin
    Carl Zeiss Meditec Inc., Dublin, California, United States
  • Footnotes
    Commercial Relationships   Luis De Sisternes, Carl Zeiss Meditec, Inc. (E); Homayoun Bagherinia, Carl Zeiss Meditec, Inc. (E); Katherine Makedonsky, Carl Zeiss Meditec, Inc. (E); Ala El Ameen, None; Giovanni Gregori, Carl Zeiss Meditec, Inc. (F), Carl Zeiss Meditec, Inc. (P); Philip Rosenfeld, Carl Zeiss Meditec, Inc. (F), Carl Zeiss Meditec, Inc. (C); Mary Durbin, Carl Zeiss Meditec, Inc. (E)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 3487. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Luis De Sisternes, Homayoun Bagherinia, Katherine Makedonsky, Ala El Ameen, Giovanni Gregori, Philip Rosenfeld, Mary K Durbin; Automated volumetric choroidal neovascularization segmentation and quantification in swept-source OCT angiography using machine learning. Invest. Ophthalmol. Vis. Sci. 2019;60(9):3487.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : Volumetric characteristics of choroidal neovascularization (CNV) are difficult to quantify in swept-source OCT angiography (SS-OCTA) due to difficulties making volumetric manual annotations. We propose a method to automatically segment and quantify CNVs in a volumetric manner in SS-OCTA data, enabling the automated characterization of CNV activity.

Methods : Retinal volumes of either a 3x3mm or a 6x6mm field of view were collected using a PLEX® Elite 9000 with AngioPlex OCT Angiography (ZEISS, Dublin, CA). Projection artifacts were corrected in a volumetric manner throughout the SS-OCTA data. A set of 31 features was extracted from the angiography data describing intensity, textural, and location angiography characteristics for each pixel contained between the internal limiting membrane and 60 microns below Bruch’s membrane. This set of features was used to train a machine learning model that identifies which pixels are contained within the CNV volume, producing volumetric segmentation of CNV. As a result of the volumetric segmentation procedure, a set of 10 characteristics were automatically extracted that describe the CNV volume, area, density, and invasiveness (intrusion within the retina). The model was tested in a separate set of SS-OCTA volumes to verify its accuracy.

Results : We processed 42 SS-OCTA volumes from eyes with CNV and 20 without CNV. 14 CNV and 2 non-CNV eyes were used as training volumes and were manually annotated using multiple slabs (topographic views) to learn the image characteristics of CNV and CNV-absent regions within the volumes (a total of 357 slabs annotated). Our method correctly identified CNV eyes from eyes without CNV with 100% accuracy in the tested eyes. Comparison of the automated CNV segmentation outlines with manual markings in CNV eyes yielded a median sensitivity of 0.83 and specificity of 0.94. Example result is shown in Figure 1. The features extracted from the automated segmentations provide a comprehensive characterization of CNV within the SS-OCTA volume that would be very difficult to obtain manually.

Conclusions : Our novel method provides a set of volumetric characteristics of CNV within SS-OCTA data that can be computed automatically and seems very promising to monitor CNV activity in patients over time.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.

 

Figure 1. Visualization of segmentation results and example features in a CNV case.

Figure 1. Visualization of segmentation results and example features in a CNV case.

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×