Investigative Ophthalmology & Visual Science Cover Image for Volume 63, Issue 7
June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Exploring Retinal Vascular Morphology via A Deep Learning Pipeline
Author Affiliations & Notes
  • Yukun Zhou
    Centre for Medical Image Computing, University College London, London, United Kingdom
    NIHR Biomedical Research Centre, Moorfields Eye Hospital NHS Foundation Trust, London, United Kingdom
  • Siegfried Wagner
    NIHR Biomedical Research Centre, Moorfields Eye Hospital NHS Foundation Trust, London, United Kingdom
    Institute of Ophthalmology, University College London, London, United Kingdom
  • Mark Chia
    NIHR Biomedical Research Centre, Moorfields Eye Hospital NHS Foundation Trust, London, United Kingdom
    Institute of Ophthalmology, University College London, London, United Kingdom
  • Peter Woodward-Court
    NIHR Biomedical Research Centre, Moorfields Eye Hospital NHS Foundation Trust, London, United Kingdom
    Institute of Health Informatics, University College London, London, United Kingdom
  • Daniel Alexander
    Centre for Medical Image Computing, University College London, London, United Kingdom
    Department of Computer Science, University College London, London, United Kingdom
  • Pearse Andrew Keane
    NIHR Biomedical Research Centre, Moorfields Eye Hospital NHS Foundation Trust, London, United Kingdom
    Institute of Ophthalmology, University College London, London, United Kingdom
  • Footnotes
    Commercial Relationships   Yukun Zhou None; Siegfried Wagner None; Mark Chia None; Peter Woodward-Court None; Daniel Alexander None; Pearse Keane DeepMind, Roche, Novartis, Apellis, BitFount, Code C (Consultant/Contractor), Big Picture Medical, Code O (Owner), Heidelberg Engineering, Topcon, Allergan, Bayer, Code R (Recipient)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 194 – F0041. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Yukun Zhou, Siegfried Wagner, Mark Chia, Peter Woodward-Court, Daniel Alexander, Pearse Andrew Keane; Exploring Retinal Vascular Morphology via A Deep Learning Pipeline. Invest. Ophthalmol. Vis. Sci. 2022;63(7):194 – F0041.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Retinal vascular morphology provides valuable information for both ophthalmic disease and systemic disease (termed 'oculomics'), e.g., atherosclerosis and diabetes mellitus, in a rapid and non-invasive way. To help recognise high risk cases of ophthalmic and systemic disease through observing the changes of retinal vascular morphology, we propose a deep learning pipeline to automatically analyse the vascular morphology (AutoMorph) which measures 12 kinds of metrics, such as vessel calibre and tortuosity.

Methods : AutoMorph consists of four functional modules: image pre-processing, image quality grading, anatomical segmentation, including binary vessel, artery/vein, and optic disc/cup segmentation, and vascular morphology feature measurement. Image quality grading and anatomical segmentation use the most recent deep learning techniques and are trained on 12 public datasets, such as DRIVE, STARE, and CHASEDB1. We employ model ensemble strategy to achieve robust results and analyse the prediction confidence to rectify false gradable cases in image quality grading. We quantitatively validate module performance on 6 external publicly available datasets including EyePACS image quality dataset (EyePACS-Q), DDR, AV-WIDE, DR-HAGIS, IOSTAR-AV, and IDRID datasets.

Results : The EfficientNet-b4 architecture used in the image grading module achieves comparable performance to the state-of-the-art method in EyePACS-Q, with an F1-score of 0.86. The confidence analysis reduces 76% of false gradable images. The binary vessel segmentation module achieves an F1-score of 0.73 on AV-WIDE and 0.78 on DR-HAGIS. The artery/vein module scores 0.66 on IOSTAR-AV, and optic disc/cup module achieves 0.94 in disc segmentation in IDRID. These results verify that AutoMorph performs well in external validation, being quantitatively on par or even better than some recent work in internal validation.

Conclusions : AutoMorph performs well even when the external validation data shows significant difference to the training data, e.g., validation on ultra-wide field retinal photography. The fully automatic pipeline integrates recent technical work to facilitate 'oculomics' research.

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.

×