June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Texture-transformer-based deep-learning (DL) network for enhancing image-quality of OCT-angiography images with lower-resolution
Author Affiliations & Notes
  • Carol Yim-lui Cheung
    Ophthalmology and Visual Sciences, The Chinese University of Hong Kong, Hong Kong
  • Hao Chen
    Computer Science and Engineering, The Hong Kong University of Sciences and Technology, Hong Kong
  • Yuyan RUAN
    Electrical Engineering, City University of Hong Kong, Hong Kong
  • Weiwen Zhang
    Mathematics, The Hong Kong University of Sciences and Technology, Hong Kong
  • Dawei Gabriel YANG
    Ophthalmology and Visual Sciences, The Chinese University of Hong Kong, Hong Kong
  • Footnotes
    Commercial Relationships   Carol Cheung None; Hao Chen None; Yuyan RUAN None; Weiwen Zhang None; Dawei YANG None
  • Footnotes
    Support  none
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 215 – F0062. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Carol Yim-lui Cheung, Hao Chen, Yuyan RUAN, Weiwen Zhang, Dawei Gabriel YANG; Texture-transformer-based deep-learning (DL) network for enhancing image-quality of OCT-angiography images with lower-resolution. Invest. Ophthalmol. Vis. Sci. 2022;63(7):215 – F0062.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Despite quantitative diabetic macular ischemia (DMI) assessment using optical coherence tomography-angiography (OCT-A) is gradually recognized, most of OCT-A studies only focus on using small-of-view with high-resolution (HR) scanning protocol to ensure the reliability of measurement. We aim to develop a DL network to reconstruct 6mm×6mm OCT-A images with image-quality enhancement to tackle the issue of “resolution trade-off” between field-of-view, image-quality and scanning time.

Methods : A novel texture-transformer-based DL network was built for reconstruction task for 6mm×6mm (320A-scan/320B-scan) fovea-center OCT-A images (Fig 1) acquired from a swept-source OCT (Triton DRI-OCT, Topcon, Inc., Japan). 296 eyes from 158 individuals with diabetes were used for training and primary validation (8:2) with each eye consists of one 3mmx3mm and one 6mm×6mm fovea-center OCT-A images. Structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) were used to compare the image-quality before and after reconstruction. To test the relationship between OCT-A metrics and diabetic retinopathy (DR) severity on the reconstructed images. We used a non-overlapping dataset (no DR: 22; mild DR:20; moderate DR:18; severe DR or proliferative DR:18) with each eye consists of 5 3mmx3mm (320A-scan/320B-scan, 1 fovea-center and 4 at parafoveal regions) and 1 6mm×6mm (320A-scan/320B-scan, fovea-center) OCT- A images. The 5 3mmx3mm OCTA images were then stitched as montage to provide a HR 6mm×6mm image serving as the “ground-truth” to the reconstructed images.

Results : After image reconstruction, better visibility of retinal capillaries are appreciated (Fig 2). The SSIM (SSIMoriginal = 0.670 vs. SSIMreconstructed = 0.500, p<0.001) and PSNR (PSNR original=19.71 vs. PSNRreconstructed=17.41, p< 0.001) of the reconstructed images was significantly higher than that of the original images. In the non-overlapping dataset, we measured foveal avascular zone (FAZ) area, FAZ circularity, vessel density on both the ground-truth and reconstructed images. We found that the associations of the OCT-A metrics with DR severity between “ground-truth” and reconstructed images were large identical (i.e. similar R2 values).

Conclusions : The proposed DL network can enhance the image-quality among OCT-A images with lower resolution and maintain the accuracy of quantitative OCT-A measurements.

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.

×