June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Background noise-resolved and enhanced OCT angiography using an end-to-end convolutional neural network
Author Affiliations & Notes
  • Min Gao
    Oregon Health & Science University, Portland, Oregon, United States
  • Yukun Guo
    Oregon Health & Science University, Portland, Oregon, United States
  • Tristan Hormel
    Oregon Health & Science University, Portland, Oregon, United States
  • Jie Wang
    Oregon Health & Science University, Portland, Oregon, United States
  • Thomas S Hwang
    Oregon Health & Science University, Portland, Oregon, United States
  • Yali Jia
    Oregon Health & Science University, Portland, Oregon, United States
  • Footnotes
    Commercial Relationships   Min Gao None; Yukun Guo None; Tristan Hormel None; Jie Wang Visionix/Optovue, Code P (Patent); Thomas Hwang None; Yali Jia Visionix/Optovue, Code P (Patent), Optos, Code P (Patent)
  • Footnotes
    Support  This work was supported by grant National Institutes of Health (R01 EY027833, R01 EY024544, R01 EY031394, T32 EY023211, UL1TR002369, P30 EY010572); Unrestricted Departmental Funding Grant and Dr. H. James and Carole Free Catalyst Award from Research to Prevent Blindness (New York, NY); Bright Focus Foundation (G2020168).
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 2365. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Min Gao, Yukun Guo, Tristan Hormel, Jie Wang, Thomas S Hwang, Yali Jia; Background noise-resolved and enhanced OCT angiography using an end-to-end convolutional neural network. Invest. Ophthalmol. Vis. Sci. 2023;64(8):2365.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To denoise optical coherence tomographic angiography (OCTA) images using an end-to-end deep learning-based method and investigate the impact of background noise on the measurement of biomarkers in OCTA.

Methods : In this study, eight 3×3-mm OCTA scans with a 400×400 sampling density from 52 participants (34 age-related macular degeneration, 18 diabetic retinopathy) were obtained using a commercial 120-kHz spectral-domain OCT system (Solix; Visionix/Optovue, Inc., CA, USA). Four pairs of the orthogonal scans (x-fast and y-fast) were registered to generate four motion-free volumes, and then these four volumes were registered and merged to obtain a high-definition volume. Projection-resolved OCTA algorithm removed projection artifacts in the voxel. The superficial vascular complex, intermediate capillary plexus, deep capillary plexus angiograms, and inner retinal angiograms were generated by maximum projection of the OCTA signal within the respective plexus. We proposed an end-to-end convolutional neural network to denoise and enhance enface angiograms. This method takes the OCTA images with various noise levels as inputs and outputs denoised and enhanced vascular network. The Mann-Whitney U-test was applied to compare the original and output angiograms using noise intensity in the foveal avascular zone (FAZ), RMS contrast, and vascular connectivity. The vessel density and non-perfusion area (NPA) in FAZ were measured in the original and denoised angiograms.

Results : Compared to the original angiograms, the noise intensity of denoised angiograms was significantly reduced (mean ± std, 137.7 ± 64.2 vs. 0.9 ± 1.9, P<0.001); the contrast of denoised angiograms was significantly enhanced (mean ± std, 52.7 ± 2.8 vs. 71.4 ± 4.1, P<0.001); the connectivity of denoised angiograms was significantly improved (mean ± std, 0.95 ± 0.01 vs. 0.98 ± 0.01, P<0.001) [Fig.1]. Compared to the original angiograms, the accuracy of NPA segmentation in FAZ of the output angiograms was significantly increased (mean ± std, 95.4% ± 10.6% vs. 99.4% ± 2.3%, P<0.001) [Fig.2 (A2-D2)]; the vessel density of FAZ in the denoised angiograms was significantly reduced (mean ± std, 5.3×10-2 ± 2.4 ×10-2 vs. 6.3×10-3±1.2×10-2, P<0.001) [Fig.2 (A3-D3)].

Conclusions : Background noise-resolved and enhanced OCTA using deep learning can improve the accuracy and reliability in the measurement of biomarkers, such as NPA and vessel density.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

 

×
×

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.

×