July 2019
Volume 60, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2019
Automated double-layer-sign detection in patients with age-related macular degeneration using deep learning algorithms applied to OCT scans
Author Affiliations & Notes
  • Yuxuan Cheng
    Bioengineering , University of Washington, Seattle, Washington, United States
  • Zhongdi Chu
    Bioengineering , University of Washington, Seattle, Washington, United States
  • Qinqin Zhang
    Bioengineering , University of Washington, Seattle, Washington, United States
  • Yingying Shi
    Department of Ophthalmology, Bascom Palmer Eye Institute, Miami, Florida, United States
  • Giovanni Gregori
    Department of Ophthalmology, Bascom Palmer Eye Institute, Miami, Florida, United States
  • Philip J. Rosenfeld
    Department of Ophthalmology, Bascom Palmer Eye Institute, Miami, Florida, United States
  • Ruikang K Wang
    Bioengineering , University of Washington, Seattle, Washington, United States
  • Footnotes
    Commercial Relationships   Yuxuan Cheng, None; Zhongdi Chu, None; Qinqin Zhang, None; Yingying Shi, None; Giovanni Gregori, None; Philip Rosenfeld, None; Ruikang Wang, None
  • Footnotes
    Support  Research supported by grants from Carl Zeiss Meditec, Inc. (Dublin, CA), the Salah Foundation, the National Eye Institute Center Core Grant (P30EY014801) and Research to Prevent Blindness (unrestricted Grant) to the Department of Ophthalmology, University of Miami Miller School of Medicine.
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 1865. doi:https://doi.org/
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Yuxuan Cheng, Zhongdi Chu, Qinqin Zhang, Yingying Shi, Giovanni Gregori, Philip J. Rosenfeld, Ruikang K Wang; Automated double-layer-sign detection in patients with age-related macular degeneration using deep learning algorithms applied to OCT scans. Invest. Ophthalmol. Vis. Sci. 2019;60(9):1865. doi: https://doi.org/.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : The presence of a double-layer sign (DLS) is predictive of subclinical type 1 macular neovascularization(MNV). Manual identification of DLS from 3D OCT scans is effective, but time-consuming and not clinically practical. The purpose of this study is to develop an automated deep learning algorithm to identify the DLS in 3D OCT scans of patient suspects of type 1 MNV.

Methods : Swept-source OCT and OCT angiography (OCTA) datasets of 40 non-exudative AMD eyes with subclinical MNV, 20 normal eyes, and 20 eyes with drusen (500 OCT B-scans per eye) were used to train and test a two-stage MNV detection model (Figure 1.a&b). The first stage model was designed to detect layer separation of Bruch’s membrane and the retinal pigment epithelium using a convolutional neural network. The second stage model was designed to take all positive B-scans from stage one and differentiate true MNV from drusen. In total, 16000 B-scans were used in stage one training and testing (26.3% MNV, 73.6% normal) For stage two, 30000 B-scans was used for training and testing (14% MNV, 16% drusen and 70% normal). All regions with MNV and drusen were manually labeled on en faceOCTA images (Figure 1.d).

Results : The stage one model achieved an accuracy of 93.2% and a sensitivity of 95.1% on a test dataset, and the stage two model achieved an accuracy of 75.2% and a sensitivity of 82.9%. The processing time was 2.76 seconds per OCT B-scan with the Titan Xp GPU. After processing, the two-stage model could provide a color-coded B-scan (Figure 1.f) to indicate locations suspicious for MNV and could also generate an en face color map (Figure 1.e) that indicated the presence of possible MNV.

Conclusions : Our deep learning model can identify MNV on OCT B-scans by detecting DLS automatically with good sensitivity and accuracy. This approach could be a clinically useful tool for detecting subclinical MNV from routine structural OCT scans before spending the time and effort to search for non-exudative MNV on B-scans or boundary-specific segmentation using OCT angiography scans.

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

 

Figure 1. (a) Stage I workflow for RPE and Bruch’s membrane separation detection. (b) Stage II workflow for MNV detection (c) en-face projection of MNV produced by OCTA and (d) its binary mask as the target for training. (e) The en-face prediction of the DLS. (f) The region where presents the double layer sign in B-scan.

Figure 1. (a) Stage I workflow for RPE and Bruch’s membrane separation detection. (b) Stage II workflow for MNV detection (c) en-face projection of MNV produced by OCTA and (d) its binary mask as the target for training. (e) The en-face prediction of the DLS. (f) The region where presents the double layer sign in B-scan.

×
×

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.

×