June 2021
Volume 62, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2021
Monitoring GA lesion size on OCT using automated deep learning-based image segmentation in the FILLY phase II clinical trial
Author Affiliations & Notes
  • Dmitrii Lachinov
    Dept. of Ophthalmology and Optometry, Medical University of Vienna, Austria
  • Julia Mai
    Dept. of Ophthalmology and Optometry, Medical University of Vienna, Austria
  • Gregor Sebastian Reiter
    Dept. of Ophthalmology and Optometry, Medical University of Vienna, Austria
  • Sophie Riedl
    Dept. of Ophthalmology and Optometry, Medical University of Vienna, Austria
  • Wolf-Dieter Vogl
    Dept. of Ophthalmology and Optometry, Medical University of Vienna, Austria
  • Ursula Schmidt-Erfurth
    Dept. of Ophthalmology and Optometry, Medical University of Vienna, Austria
  • Hrvoje Bogunovic
    Dept. of Ophthalmology and Optometry, Medical University of Vienna, Austria
  • Footnotes
    Commercial Relationships   Dmitrii Lachinov, None; Julia Mai, None; Gregor Reiter, None; Sophie Riedl, None; Wolf-Dieter Vogl, None; Ursula Schmidt-Erfurth, Apellis Pharmaceuticals (F), Genentech (C), Heidelberg Engineering (C), Kodiak (C), Novartis (C), RetInSight (C), Roche (C); Hrvoje Bogunovic, Apellis Pharmaceuticals (F)
  • Footnotes
    Support  Apellis Pharmaceuticals
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 239. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Dmitrii Lachinov, Julia Mai, Gregor Sebastian Reiter, Sophie Riedl, Wolf-Dieter Vogl, Ursula Schmidt-Erfurth, Hrvoje Bogunovic; Monitoring GA lesion size on OCT using automated deep learning-based image segmentation in the FILLY phase II clinical trial. Invest. Ophthalmol. Vis. Sci. 2021;62(8):239.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : In patients with geographic atrophy (GA) secondary to age-related macular degeneration (AMD), the rate of progression is variable. To evaluate the potential of artificial intelligence (AI) for active monitoring of GA progression, we trained and evaluated an automated deep learning-based image segmentation algorithm to detect and measure the size of retinal pigment epithelium (RPE)-loss on OCT scans.

Methods : OCT scans (512x49x496 voxels, Heidelberg Engineering) of study eyes of patients with complete RPE and outer retinal atrophy (cRORA) were evaluated. Patients were enrolled in the FILLY phase II clinical trial to study the effect of pegcetacoplan (APL-2), an investigational therapy targeting complement C3. The scans were first manually annotated for the presence of cRORA at an A-scan level, delineating the GA area. With the manual annotations as the reference, a deep learning image segmentation method using a 3D-to-2D convolutional neural network (CNN) was then trained to automatically segment a topographic 2D cRORA area on a 3D-OCT scan and its performance was evaluated using a five-fold cross-validation.

Results : 115 study eyes were included in the analysis that completed a one year follow-up and had a sufficient image quality to allow manual annotation of RPE-loss at baseline and year 1. The segmentation performance was evaluated as a precision, recall and Dice score (DSC), corresponding to the overlap between the reference and the automated segmentation. For the set of baseline scans, a mean DSC was 0.91, precision was 0.90, and recall was 0.94. For the year 1 scans, the mean DSC was 0.93, precision was 0.91, and recall was 0.95. Evaluation limited to the differential one-year growth area, resulted in the mean DSC of 0.57, precision of 0.49, and recall of 0.72.

Conclusions : Fully automated image segmentation of RPE loss on OCT is able to localize and delineate the GA area with high accuracy. Automated AI-based analysis was able to reproduce the findings of the phase II clinical trial complementing the results of centralized reading. The results of this pilot study represent a promising step toward AI-based clinical decision support tools that can monitor GA progression and therapeutic response, once the treatments for GA secondary to AMD become available.

This is a 2021 ARVO Annual Meeting abstract.

×
×

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.

×