A Structure-Consistency GAN for Unpaired AS-OCT Image Inpainting

Guanhua Bai, Sanqian Li, He Zhang, Risa Higashita, Jiang Liu, Jie Li, Meng Zhang

Research output: Chapter in Book/Conference proceedingConference contributionpeer-review


Anterior segment optical coherence tomography (AS-OCT) is a crucial imaging modality in ophthalmology, providing valuable insights into corneal pathologies. However, during AS-OCT imaging, intense signals in highly reflective regions can easily lead to saturation effects, resulting in pronounced stripes across the cornea. It compromises the image visual quality and impacts automated ophthalmic analysis. To address this issue, we propose an unsupervised Structure-Consistency Generative Adversarial Network (SC-GAN) that captures the underlying semantic structural knowledge in both the spatial domain and frequency space within the generative model. This strategy aims to mitigate the influence of bright stripes and restore corneal structural details in AS-OCT images. Specifically, SC-GAN introduces a stripe perceptual loss to extract visual representations by utilizing the perceptual similarity between striped and stripe-free images. Moreover, Fourier feature mapping is adopted to learn high-frequency information, thereby achieving crucial structure consistency. The experimental results demonstrate that the proposed SC-GAN can removes stripes while preserving crucial corneal structures, surpassing the competing algorithms. Furthermore, we validate the benefits of SC-GAN in the corneal segmentation task.

Original languageEnglish
Title of host publicationOphthalmic Medical Image Analysis - 10th International Workshop, OMIA 2023, Held in Conjunction with MICCAI 2023, Proceedings
EditorsBhavna Antony, Hao Chen, Huihui Fang, Huazhu Fu, Cecilia S. Lee, Yalin Zheng
PublisherSpringer Science and Business Media Deutschland GmbH
Number of pages10
ISBN (Print)9783031440120
Publication statusPublished - 2023
Externally publishedYes
Event10th International Workshop on Ophthalmic Medical Image Analysis, OMIA-X 2023 - Vancouver, Canada
Duration: 12 Oct 202312 Oct 2023

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume14096 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference10th International Workshop on Ophthalmic Medical Image Analysis, OMIA-X 2023


  • AS-OCT
  • GAN
  • Inpainting
  • structural consistency

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science (all)


Dive into the research topics of 'A Structure-Consistency GAN for Unpaired AS-OCT Image Inpainting'. Together they form a unique fingerprint.

Cite this