June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Accuracy of Deep-Learning-Derived OCT Retinal Layer Segmentation in the Alabama Study on Early Age-Related Macular Degeneration 2 (ALSTAR2)
Author Affiliations & Notes
  • Sohaib Fasih-Ahmad
    Doheny Eye Institute, Pasadena, California, United States
  • Ziyuan Wang
    Doheny Eye Institute, Pasadena, California, United States
  • Zubin Mishra
    Doheny Eye Institute, Pasadena, California, United States
  • Charles Vatanatham
    Doheny Eye Institute, Pasadena, California, United States
  • Anjal M Jain
    Doheny Eye Institute, Pasadena, California, United States
  • Anushika Ganegoda
    Doheny Eye Institute, Pasadena, California, United States
  • Zhihong Jewel Hu
    Doheny Eye Institute, Pasadena, California, United States
  • Mark Clark
    The University of Alabama at Birmingham Department of Ophthalmology and Visual Sciences, Birmingham, Alabama, United States
  • Cynthia Owsley
    The University of Alabama at Birmingham Department of Ophthalmology and Visual Sciences, Birmingham, Alabama, United States
  • Christine A Curcio
    The University of Alabama at Birmingham Department of Ophthalmology and Visual Sciences, Birmingham, Alabama, United States
  • Srinivas R Sadda
    Doheny Eye Institute, Pasadena, California, United States
    Department of Ophthalmology, University of California Los Angeles, Los Angeles, California, United States
  • Footnotes
    Commercial Relationships   Sohaib Fasih-Ahmad None; Ziyuan Wang None; Zubin Mishra None; Charles Vatanatham None; Anjal Jain None; Anushika Ganegoda None; Zhihong Jewel Hu None; Mark Clark None; Cynthia Owsley MacuLogix, Code P (Patent); Christine Curcio Genentech/Hoffman LaRoche, Regeneron, Code C (Consultant/Contractor), MacRegen, Code I (Personal Financial Interest); Srinivas Sadda Amgen, Allergan, Genentech/Roche, Iveric, Oxurion, Novartis, Regeneron, Bayer, 4DMT, Centervue, Heidelberg, Optos, Merck, Apellis, Astellas, Code C (Consultant/Contractor), Carl Zeiss Meditec, Nidek, Code R (Recipient), Nidek, Topcon, Heidelberg, Carl Zeiss Meditec, Optos, Centervue, Code R (Recipient)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2075 – F0064. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Sohaib Fasih-Ahmad, Ziyuan Wang, Zubin Mishra, Charles Vatanatham, Anjal M Jain, Anushika Ganegoda, Zhihong Jewel Hu, Mark Clark, Cynthia Owsley, Christine A Curcio, Srinivas R Sadda; Accuracy of Deep-Learning-Derived OCT Retinal Layer Segmentation in the Alabama Study on Early Age-Related Macular Degeneration 2 (ALSTAR2). Invest. Ophthalmol. Vis. Sci. 2022;63(7):2075 – F0064.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Manual segmentation of early AMD features and associated retinal layers in SD-OCT images is tedious and time-consuming. Using only 5 OCT volumes of the ALSTAR cohort, we trained our existing deep-learning derived, graph-based algorithm to segment the outer-retinal layers in eyes with early AMD. Our goal was to evaluate the accuracy of this algorithm in ALSTAR2 cohort compared with manual segmentation.

Methods : Subjects ≥ 60 years old with healthy eyes or early AMD in at least one eye were enrolled in the Alabama Study on Early Age-Related Macular Degeneration 2 (ALSTAR2). Eye-tracked SD-OCT volumes (8.6x7.2mm, ART > 9, fovea centered) were acquired with Spectralis HRA + OCT2. The outer-retinal layers [inner segment myoid (ISM) (defined to span from the external limiting membrane to ellipsoid zone (EZ) center line), photoreceptor outer segment (EZ center line to outer surface of interdigitation zone), subretinal drusenoid deposits (SDD), retinal pigment epithelium (RPE), drusen, full choroid] were segmented manually by expert graders and fully automatically by our algorithm. Based on these segmentations, mean thickness, total volume, and total area of the 6 layers were computed in 9 ETDRS subfields, and mean measurement difference and correlation coefficients were calculated.

Results : The study cohort included 400 eyes (140 early AMD, 260 normal). The mean difference and correlation coefficients between segmentation methods are shown in Table 1. In general, mean thickness differences for most layers were <1µm (choroid by ~5-7µm) with a significant correlation between the automated and manual measurements for ISM, photoreceptor outer segments, RPE, and choroid layers. Correlations were not as good for drusen and SDD layers.

Conclusions : A deep-learning derived algorithm has promise in segmenting hyporeflective bands of ISM, photoreceptor outer segments, RPE, and choroid in normal and early AMD eyes. Further algorithm training using more OCT volumes with drusen and SDD lesions may be required to improve algorithm performance, and further statistical analysis is planned.

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

 

Note: NaN indicates undefined correlation due to lack of variance
* p > 0.05
Table 1. Association of mean thickness, total area, and volume of outer-retinal layers between automated and manually segmented OCT images in 9 ETDRS subfields

Note: NaN indicates undefined correlation due to lack of variance
* p > 0.05
Table 1. Association of mean thickness, total area, and volume of outer-retinal layers between automated and manually segmented OCT images in 9 ETDRS subfields

 

Figure 1. Ex. of segmentations of half of a B-scan

Figure 1. Ex. of segmentations of half of a 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.

×