August 2019
Volume 60, Issue 11
Free
ARVO Imaging in the Eye Conference Abstract  |   August 2019
Eliminating retinal vessel shadows in en face choroidal OCT via generative adversarial networks
Author Affiliations & Notes
  • Jianlong Yang
    Cixi Institute of Biomedical Engineering, Chinese Academy of Sciences, Ningbo, China
  • Huihong Zhang
    Cixi Institute of Biomedical Engineering, Chinese Academy of Sciences, Ningbo, China
  • Kang Zhou
    Cixi Institute of Biomedical Engineering, Chinese Academy of Sciences, Ningbo, China
    School of Information Science and Technology, ShanghaiTech University, China
  • Fei Li
    Zhongshan Ophthalmic Center, State Key Laboratory of Ophthalmology, Sun Yat-sen University, China
  • Ce Zheng
    Ophthalmology of Shanghai Children’s Hospital, China
  • Yan Hu
    Cixi Institute of Biomedical Engineering, Chinese Academy of Sciences, Ningbo, China
  • Xiulan Zhang
    Zhongshan Ophthalmic Center, State Key Laboratory of Ophthalmology, Sun Yat-sen University, China
  • Jiang Liu
    Cixi Institute of Biomedical Engineering, Chinese Academy of Sciences, Ningbo, China
    Department of Computer Science and Engineering, Southern University of Science and Technology, China
  • Footnotes
    Commercial Relationships   Jianlong Yang, None; Huihong Zhang, None; Kang Zhou, None; Fei Li, None; Ce Zheng, None; Yan Hu, None; Xiulan Zhang, None; Jiang Liu, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science August 2019, Vol.60, PB091. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Jianlong Yang, Huihong Zhang, Kang Zhou, Fei Li, Ce Zheng, Yan Hu, Xiulan Zhang, Jiang Liu; Eliminating retinal vessel shadows in en face choroidal OCT via generative adversarial networks. Invest. Ophthalmol. Vis. Sci. 2019;60(11):PB091.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : In the OCT imaging of the eye, the morphology of the choroidal vasculature is influenced by the vessel shadows from the retinal layers. Recently, an attenuation compensation (AC) approach to enhance the B-frames was proposed for eliminating the retinal vessel shadows. From the en face OCT view, the shadow removal can be seen as an inpainting task, which refers to repairing the shadow-contaminated regions of the choroid. In this work, we propose to use a two-stage generative adversarial network (GAN) for this task and compare its performance with that of a traditional coherence transport inpainting (CTI) method.

Methods : The OCT data sets we used were 6×6 mm2 volumetric scans from a Topcon 100-kHz swept source system. We employed a two-stage GAN inpainting architecture as shown in Fig. 1. The retinal vessel mask was extracted from the RPE layer. It was fed into the network together with the input en face choroidal OCT and its Canny edge map. The first stage of the GAN was used to connect the vessel edges at the locations of the shadow mask. Then the connected edge map combining with the en face OCT image were used in the second stage of the GAN to finish the shadow removal task. Our generators used Perceptual GAN architecture and the discriminators used PatchGAN architecture. Two fine-tune layers were added at each stage to fit our gray scale images with the pre-trained RGB models.

Results : As shown in Fig. 2, the AC method is capable of removing small-vessel and capillary shadows while leaves the large-vessel shadows unaffected. The inpainting-based methods can entirely eliminate the shadows, but the CTI method brings smeary and blurry patterns around the inpainting regions. The proposed GAN-based method, on the other hand, is able to avoid generating artificial patterns and preserve the shape, connectivity, and contrast of the choroidal vessels.

Conclusions : The proposed GAN-based method outperforms the existing AC and CTI methods in both the vessel elimination and preserving the morphology of the choroidal vasculature. It will benefit the quantification of the OCT-based analysis and diagnoses in ophthalmology.

This abstract was presented at the 2019 ARVO Imaging in the Eye Conference, held in Vancouver, Canada, April 26-27, 2019.

 

Architecture of the two-stage generative adversarial network for the shadow removal task.

Architecture of the two-stage generative adversarial network for the shadow removal task.

 

Comparison of different methods for eliminating the retinal vessel shadows. The retinal vessel masks generated from the en face RPE image are presented here for indicating the locations of the shadows.

Comparison of different methods for eliminating the retinal vessel shadows. The retinal vessel masks generated from the en face RPE image are presented here for indicating the locations of the shadows.

×
×

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.

×