July 2019
Volume 60, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2019
Early Photoreceptor Alterations in OCT for Quantitative Prediction of Expansion and New Foci of Geographic Atrophy using Deep Learning
Author Affiliations & Notes
  • Zhihong Hu
    Retina, Doheny Eye Institute, Arcadia, California, United States
  • Ziyuan Wang
    Retina, Doheny Eye Institute, Arcadia, California, United States
  • Srinivas R Sadda
    Retina, Doheny Eye Institute, Arcadia, California, United States
    Ophthalmology, University of California, Los Angeles, Los Angeles, California, United States
  • Footnotes
    Commercial Relationships   Zhihong Hu, None; Ziyuan Wang, None; Srinivas Sadda, Allergan (F), Allergan (C), Carl Zeiss Meditec (F), Centervue (C), Genentech (F), Genentech (C), Heidelberg (C), Iconic (C), Novartis (C), Optos (F), Optos (C), Oxurion (C), Topcon (R)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 1535. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Zhihong Hu, Ziyuan Wang, Srinivas R Sadda; Early Photoreceptor Alterations in OCT for Quantitative Prediction of Expansion and New Foci of Geographic Atrophy using Deep Learning. Invest. Ophthalmol. Vis. Sci. 2019;60(9):1535.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Most existing quantitative studies of geographic atrophy (GA) associated with age-related macular degeneration (AMD) using spectral domain optical coherence tomography (SD-OCT), utilize the characteristic of light signal penetration into the choroid (termed “hyper-transmission”) due to the disappearance of the retinal pigment epithelium (RPE) layer. However, early photoreceptor alterations on OCT can be observed before the manifestation of hyper-transmission, which may foreshadow future atrophy progression (expansion or new foci), and may be useful for identifying patients who may be at risk for more rapid progression. As a pilot investigation, the purpose of this study was to utilize deep learning approaches to identify these early predictive photoreceptor alterations on OCT.

Methods : The study cohort (from anonymized reading center datasets) consisted of 70 eyes of 70 patients with AMD with fundus autofluorescence (FAF; Heidelberg Spectralis) and SD-OCT (Zeiss Cirrus OCT; 6x6 mm, 512x128 macular cube) data collected over 12 months. GA has been commonly measured on FAF as an endpoint or outcome measure. The areas of early photoreceptor alterations were segmented from OCT en face maps at baseline using a state-of-the-art semantic deep learning segmentation algorithm u-net trained against manually delineated GA at a follow-up visit of month 12 from FAF images. The en face OCT maps were automatically generated from the layer of external limiting membrane (ELM) and inner RPE. The algorithm-detected photoreceptor alteration regions from OCT en face maps at baseline were compared with the manually delineated GA regions on FAF at month 12.

Results : A high degree of intraclass correlation (ICC) was found between the algorithm-detected photoreceptor alteration regions on OCT at baseline and manually-delineated GA regions on FAF at month 12. The average measure of the ICCs was 0.907 with a 95% confidence interval from 0.803 to 0.956 (F(29,29) = 10.397, p < 0.001).

Conclusions : Using a deep learning approach, early photoreceptor alterations could be identified on OCT in eyes with AMD, which correlated with the progression of GA into these regions 12 months later. If replicated in larger prospective cohorts, these findings may be useful in identifying subjects who may have greater risk for progression of atrophy.

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

 

 

×
×

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.

×