June 2023
Volume 64, Issue 9
Open Access
ARVO Imaging in the Eye Conference Abstract  |   June 2023
Novel artificial intelligence-assisted optical coherence tomography-angiography delineation of retinal vessels and leakage by deep learning with fluorescein angiography video training data
Author Affiliations & Notes
  • Toshinori Murata
    Ophthalmology, Shinshu University, Matsumoto, Nagano, Japan
  • Takao Hirano
    Ophthalmology, Shinshu University, Matsumoto, Nagano, Japan
  • Hideaki Mizobe
    Canon Inc, Japan
  • Shuhei Toba
    Canon Inc, Japan
  • Footnotes
    Commercial Relationships   Toshinori Murata, Bayer (R), Chugai (R), Kowa (R), Novartis (R), Santen (R), Zeiss Meditec (R); Takao Hirano, Bayer (R), Novartis (R), Santen (R), Zeiss Meditec (R); Hideaki Mizobe, CANON INC (E), CANON INC (P); Shuhei Toba, CANON INC (E), CANON INC (P)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2023, Vol.64, PP0013. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Toshinori Murata, Takao Hirano, Hideaki Mizobe, Shuhei Toba; Novel artificial intelligence-assisted optical coherence tomography-angiography delineation of retinal vessels and leakage by deep learning with fluorescein angiography video training data. Invest. Ophthalmol. Vis. Sci. 2023;64(9):PP0013.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : In the treatment planning of diabetic retinopathy (DR), fluorescein angiography (FA) has been essential for depicting the status of retinal vessels and pathological leakage, although there have been cases of allergic shock due to contrast dye. Recently, optical coherence tomography-angiography (OCTA) has gained popularity since it can delineate retinal vessels without contrast dye and the worry of allergic complications. However, a major drawback of OCTA is that it cannot depict vessel leakage, which is critical for treating DR, including diabetic macular edema. This study evaluated the ability of a novel artificial intelligence (AI)-assisted OCTA system also able to delineate the pathological leakage of retinal vessels.

Methods : High-resolution wide-field (23 mm x 20 mm) OCTA (OCT-S1, Canon, Tokyo, Japan) images were obtained from 26 patients with DR. For each recording, approximately 20000 FA training data images were prepared from a 30-45-second FA video (Spectralis HRA, Heidelberg, Germany). AI-FA-like images were generated based on OCTA and the 20000 FA images from various time points using a convolutional neural network. Then, a generative adversarial network was adopted to make AI-FA images more similar to FA images/training data.

Results : The similarity between the AI-FA like images (Fig1a), which were generated based on OCTA images (Fig1b), and the conventional FA images (Figure2) achieved an average structural similarity index measure of 0.91 (standard deviation: 0.08). Using AI-FA images, a FA video of corresponding time points was also successfully generated.

Conclusions : With this novel AI-assisted OCTA system, high-resolution images of retinal vessels and their obstructed lesions, microaneurysms, and associated pathological leakage can be safely and clearly delineated without contrast agents.

This abstract was presented at the 2023 ARVO Imaging in the Eye Conference, held in New Orleans, LA, April 21-22, 2023.

 

Artificial intelligence-generated fluorescein angiography image (a) based on a conventional optical coherence tomography-angiography image (b). Pathological leakage from retinal vessels is clearly demonstrated in addition to retinal vessels, as well as nonperfusion areas, and retinal neovascularization.

Artificial intelligence-generated fluorescein angiography image (a) based on a conventional optical coherence tomography-angiography image (b). Pathological leakage from retinal vessels is clearly demonstrated in addition to retinal vessels, as well as nonperfusion areas, and retinal neovascularization.

 

Conventional fluorescein angiography (FA) image used as training data to obtain the artificial intelligence-generated FA image in Figure 1.

Conventional fluorescein angiography (FA) image used as training data to obtain the artificial intelligence-generated FA image in Figure 1.

×
×

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.

×