June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Deep learning for quality assessment of optical coherence tomography angiography images
Author Affiliations & Notes
  • Jay C Wang
    Ophthalmology & Visual Science, Yale School of Medicine, New Haven, Connecticut, United States
    Northern California Retina Vitreous Associates Inc, Mountain View, California, United States
  • Rahul Dhodapkar
    Ophthalmology & Visual Science, Yale School of Medicine, New Haven, Connecticut, United States
  • Emily Li
    Oculoplastics and Reconstructive Surgery, Johns Hopkins Medicine Wilmer Eye Institute, Baltimore, Maryland, United States
  • Kristen Harris Nwanyanwu
    Ophthalmology & Visual Science, Yale School of Medicine, New Haven, Connecticut, United States
  • Ron A Adelman
    Ophthalmology & Visual Science, Yale School of Medicine, New Haven, Connecticut, United States
  • Smita Krishnaswamy
    Genetics, Yale University, New Haven, Connecticut, United States
    Computer Science, Yale University, New Haven, Connecticut, United States
  • Footnotes
    Commercial Relationships   Jay Wang None; Rahul Dhodapkar None; Emily Li None; Kristen Nwanyanwu None; Ron Adelman None; Smita Krishnaswamy None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 202 – F0049. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Jay C Wang, Rahul Dhodapkar, Emily Li, Kristen Harris Nwanyanwu, Ron A Adelman, Smita Krishnaswamy; Deep learning for quality assessment of optical coherence tomography angiography images. Invest. Ophthalmol. Vis. Sci. 2022;63(7):202 – F0049.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Optical coherence tomography angiography (OCTA) is a relatively novel technology that continues to improve, but variation in image quality remains problematic for reliable image analysis. We developed a deep learning-based system using convolutional neural networks for automated quality assessment of optical coherence tomography (OCTA) images.

Methods : We performed a single center, retrospective study of diabetic patients from August 2017 to April 2019. OCTA imaging with the Cirrus HD-OCT 5000 AngioPlex was performed. Manual grading of image quality was performed on 8x8 mm and 6x6 mm superficial slab images by two independent graders. An image quality score was assigned based on presence of motion and segmentation artifacts, media opacity, and visibility of fine capillaries. Machine-reported signal strength was also recorded. Each image was assigned a final image quality score from 0 to 4 based on a composite assessment from both graders. A ResNet152 neural network classifier pretrained using ImageNet was trained to classify the images. Because requirements for image quality may vary depending on the clinical or research setting, two models were trained – one to identify highquality images and one to identify low-quality images.

Results : 347 scans from 134 patients were included. Our neural network models demonstrated outstanding area under the curve (AUC) metrics for both low quality image identification (AUC=0.98, 95%CI: 0.96-0.99, κ=0.88) and high quality image identification (AUC=0.99, 95%CI: 0.99-1.00, κ=0.92), significantly outperforming machine-reported signal strength (AUC=0.78, 95%CI: 0.73-0.83, κ=0.27 and AUC=0.82, 95%CI: 0.77-0.86, κ=0.52 respectively).

Conclusions : Deep convolutional neural networks featuring skip connections were successfully trained to automatically classify between gradable and ungradable OCTA images, and have been made publicly available as a resource to other physicians and scientists. As OCTA becomes more widely utilized, flexible and robust methods for image quality control will be important.

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

 

Representative examples of 8x8 mm OCTA images of the superficial capillary plexus with inage quality score of 2 (A and B), 1 (C and D), and 0 (E and F). Image artifacts displayed include blink lines (arrowheads), segmentation artifact (asterisks), and media opacity (arrows).

Representative examples of 8x8 mm OCTA images of the superficial capillary plexus with inage quality score of 2 (A and B), 1 (C and D), and 0 (E and F). Image artifacts displayed include blink lines (arrowheads), segmentation artifact (asterisks), and media opacity (arrows).

×
×

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.

×