June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Machine learning approaches to automated prediction of fibrosis development in neovascular age-related macular degeneration using optical coherence tomography images
Author Affiliations & Notes
  • Andreas Maunz
    F. Hoffmann-La Roche Ltd., Basel, Switzerland
  • Ian Lloyd Jones
    F. Hoffmann-La Roche Ltd., Basel, Switzerland
  • Jules Hernandez-Sanchez
    Roche Products Ltd, Welwyn Garden City, Hertfordshire, United Kingdom
  • Beatriz Garcia Armendariz
    F. Hoffmann-La Roche Ltd., Basel, Switzerland
  • Laura Barras
    Genentech Inc, South San Francisco, California, United States
  • Siqing Yu
    F. Hoffmann-La Roche Ltd., Basel, Switzerland
  • Footnotes
    Commercial Relationships   Andreas Maunz Roche, Inc., Code E (Employment); Ian Jones Roche, Inc., Code E (Employment); Jules Hernandez-Sanchez Roche, Inc., Code E (Employment); Beatriz Armendariz Roche, Inc., Code E (Employment); Laura Barras Genentech, Inc., Code E (Employment); Siqing Yu Roche, Inc., Code E (Employment)
  • Footnotes
    Support  F. Hoffmann-La Roche Ltd., Basel, Switzerland, provided support for the study and participated in the study design; conducting the study; and data collection, management, and interpretation
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2052 – F0041. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Andreas Maunz, Ian Lloyd Jones, Jules Hernandez-Sanchez, Beatriz Garcia Armendariz, Laura Barras, Siqing Yu; Machine learning approaches to automated prediction of fibrosis development in neovascular age-related macular degeneration using optical coherence tomography images. Invest. Ophthalmol. Vis. Sci. 2022;63(7):2052 – F0041.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To build and validate machine learning (ML) models using optical coherence tomography (OCT) volume scans for predicting the onset of fibrosis in patients with neovascular age-related macular degeneration (nAMD).

Methods : 935 OCT volume scans from 1097 treatment-naïve eyes with nAMD treated with ranibizumab 0.5 or 2.0 mg on a monthly or as-needed basis over 12 months were selected post hoc from the phase 3, randomized, multicenter HARBOR trial (NCT00891735). Retinal layers and pathological features were automatically detected in the OCT volume scans using a pretrained segmentation model. ML models were trained and evaluated using OCT and/or clinical variables (choroidal neovascularization [CNV] type assessed using fluorescein angiography [FA], baseline visual acuity, and age) in 2 approaches. For feature-based learning, 90 quantitative OCT features were automatically extracted from the segmentations. For end-to-end deep learning (DL), 2 models were trained with raw (DL-raw) and segmented (DL-seg) OCT B-scans.

Results : Binary classification of fibrosis development was assessed as area under the receiver operating characteristic curve (AUC), sensitivity, and specificity, using cross-validation. The DL-seg model predicted fibrosis development with average AUC values of 0.802, similar to the DL-raw model and feature-based learning (AUC = 0.786 for both), and was generally on par with the predictive performance of clinical variables only (AUC = 0.794). Notably, CNV type alone achieved an AUC of 0.747. Adding clinical variables did not considerably improve performance for DL-seg (from 0.802 to 0.809), but improved performance for DL-raw (from 0.786 to 0.808) and feature-based learning (from 0.786 to 0.821).

Conclusions : Fully automated models using OCT segmentation data can predict fibrosis development well. The DL-seg model, DL-raw model, and feature-based learning all achieved comparable performance in predicting fibrosis and were also comparable to the clinical variables only model, which requires invasive procedures and expert training for image interpretation (FA). These findings show that ML using OCT segmentations is a promising noninvasive approach to predicting fibrosis development in nAMD.

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.

×