June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
A novel deep learning method for choroidal-scleral junction segmentation
Author Affiliations & Notes
  • Homayoun Bagherinia
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Lars Omlor
    Corporate Research and Technology, Carl Zeiss Inc., Pleasanton, California, United States
  • Luis De Sisternes
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Hao Zhou
    University of Washington, Seattle, Washington, United States
  • Jie Lu
    University of Washington, Seattle, Washington, United States
  • Yingying Shi
    University of Miami Health System Bascom Palmer Eye Institute, Miami, Florida, United States
  • Philip J Rosenfeld
    University of Miami Health System Bascom Palmer Eye Institute, Miami, Florida, United States
  • Ruikang K Wang
    University of Washington, Seattle, Washington, United States
  • Footnotes
    Commercial Relationships   Homayoun Bagherinia Carl Zeiss Meditec, Inc., Code E (Employment); Lars Omlor Carl Zeiss Meditec, Inc., Code E (Employment); Luis De Sisternes Carl Zeiss Meditec, Inc., Code E (Employment); Hao Zhou None; Jie Lu None; Yingying Shi None; Philip Rosenfeld Carl Zeiss Meditec, Inc., Code C (Consultant/Contractor), Carl Zeiss Meditec, Inc., Code F (Financial Support); Ruikang Wang Carl Zeiss Meditec, Inc., Code C (Consultant/Contractor), Carl Zeiss Meditec, Inc., Code F (Financial Support)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2060 – F0049. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Homayoun Bagherinia, Lars Omlor, Luis De Sisternes, Hao Zhou, Jie Lu, Yingying Shi, Philip J Rosenfeld, Ruikang K Wang; A novel deep learning method for choroidal-scleral junction segmentation. Invest. Ophthalmol. Vis. Sci. 2022;63(7):2060 – F0049.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Reliable choroidal layer analysis has become an important diagnostic tool for retinal diseases. We propose a choroidal-scleral junction (CSJ) segmentation method for swept-source optical coherence tomography (SS-OCT) volumes. Our method is independent of scan field of view (FOV) and designed based on a novel deep learning (DL) algorithm.

Methods : Manually graded multi-retinal layer segmentation (MLS) was used to generate the ground truth which is defined as the region between RPE and CSJ. The inputs consist of 3 mm OCT patches extracted from five neighboring B-scans along with the attenuation corrected patches. A total of 17,722 training and 10,596 validation data from 603 3x3 mm, 6x6 mm, 9x9 mm, 12x12 mm scans acquired using PLEX® Elite 9000 (ZEISS, Dublin, CA), with eye diseases such as age-related macular degeneration (AMD), were used to train a model. A network was designed with the first half of an autoencoder (encoder) followed by a discrete cosine transform (decoder). In prediction, an uncertainty algorithm detects the low segmentation confidence regions that are replaced by interpolation to generate the complete CSJ surface. 61 6x6 mm and 61 12x12 mm scans from the same eyes with eye diseases such as AMD were used to generate choroidal thickness maps using manual and automated segmentation which are defined as the distance between the MLS Bruch’s membrane (BM) and the CSJ. The performance of the algorithm was reported using correlation between manual and automated methods for each sector of the ETDRS grid (Fig 2). Regression and Bland-Altman plots were used for 3 mm, 6 mm and 11 mm zones.

Results : Fig 1 shows five neighboring 3 mm OCT input and ground truth patches, and examples for CSJ segmentation results overlaid on OCT images for different scan FOV. Fig 2 shows the ETDRS and correlation between the manual and automated segmentation in each sector including the regression and Bland-Altman plots for the central 3 mm, 6 mm and 11 mm zones. Most ETDRS sectors show significant correlation (>0.85) between manual and automated measurements. Lower correlation within the ONH vicinity was explained by increased difficulties in the BM segmentation.

Conclusions : We proposed a novel deep learning based CSJ segmentation. The choroidal thickness maps generated by automated and manual segmentation have a strong correlation and good agreement. Automated segmentation may be a valuable diagnostic tool for retinal diseases.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

 

 

×
×

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.

×