June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Artificial intelligence to identify conventional treatment patterns in neovascular age-related macular degeneration in a real-world population
Author Affiliations & Notes
  • VIRGINIA DE SOUZA LEOLINO MARES
    Ophthalmology, Universidade Federal de Minas Gerais, Belo Horizonte, MG, Brazil
    Ophthalmology, Medizinische Universitat Wien, Wien, Wien, Austria
  • Hrvoje Bogunovic
    Ophthalmology, Medizinische Universitat Wien, Wien, Wien, Austria
  • Oliver Leingang
    Ophthalmology, Medizinische Universitat Wien, Wien, Wien, Austria
  • Daniel Barthelmes
    UniversitatsSpital Zurich, Zurich, Switzerland
  • Gregor Sebastian Reiter
    Ophthalmology, Medizinische Universitat Wien, Wien, Wien, Austria
  • Ursula Schmidt-Erfurth
    Ophthalmology, Medizinische Universitat Wien, Wien, Wien, Austria
  • Footnotes
    Commercial Relationships   VIRGINIA MARES None; Hrvoje Bogunovic None; Oliver Leingang None; Daniel Barthelmes None; Gregor Reiter RetInSight, Code F (Financial Support); Ursula Schmidt-Erfurth RetInSight, Code F (Financial Support)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 3014 – F0284. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      VIRGINIA DE SOUZA LEOLINO MARES, Hrvoje Bogunovic, Oliver Leingang, Daniel Barthelmes, Gregor Sebastian Reiter, Ursula Schmidt-Erfurth; Artificial intelligence to identify conventional treatment patterns in neovascular age-related macular degeneration in a real-world population. Invest. Ophthalmol. Vis. Sci. 2022;63(7):3014 – F0284.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Optical coherence tomography (OCT) is the main diagnostic tool to detect and monitor progression of neovascular age related macular degeneration (nAMD). The purpose of this study is to predict anti-VEGF treatment requirements in nAMD using artificial intelligence (AI) based on OCT images for identifying fluid biomarkers in a real-world cohort.

Methods : OCT (Spectralis, Heidelberg Engineering) data of treatment-naïve patients with nAMD from the Fight Retinal Blindness! in Zürich were processed at baseline, and the end of the loading dose (2 months after the first anti-VEGF injection), to predict subsequent one year treatment needs. First, intraretinal (IRF), subretinal fluid (SRF) and pigment epithelial detachment (PED) were segmented using a deep learning convolutional neural network (Vienna Fluid Monitor, RetInSight, Vienna, Austria). Second, a set of quantitative features from the segmented layers and fluid regions were computed across the three central subfields at 1mm, 3mm, and 6mm, to describe retinal pathomorphology both quantitatively and spatially. Finally, using the computed set of features, a predictive model of future treatment requirements was built using machine learning and was evaluated with a ten-fold patient-level cross-validation.

Results : Two hundred and nine eyes from 164 patients were evaluated for a one year period following the loading dose. The treatment intervals ranged from 0 to 13 weeks. 100/209 eyes had lower median (≤7) and 109/209 eyes had an upper median (≥8) number of injections. The model identified the two groups (lower and upper median) based on number of injections with a mean accuracy of 0.74 (CI) area under the curve (AUC). The amount of SRF after the loading dose and at baseline in the central-3mm area were found to be the most important predictive features (Figure 1).

Conclusions : We used AI to predict treatment requirements in nAMD by correlating fluid biomarkers with resulting therapeutical patterns based on OCT images. The potential of a personalized anti-VEGF therapy for minimizing the risk of undertreatment while improving resource management and avoiding overtreatment will have to be evaluated in respect to this current state-of-the-art concept in a prospective manner.

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

 

Figure 1: The most important imaging features from baseline or the end of loading dose (LD) for predicting treatment requirement.

Figure 1: The most important imaging features from baseline or the end of loading dose (LD) for predicting treatment requirement.

×
×

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.

×