Article Text

Download PDFPDF
Creating realistic anterior segment optical coherence tomography images using generative adversarial networks
  1. Jad F Assaf1,2,
  2. Anthony Abou Mrad1,
  3. Dan Z Reinstein3,4,5,6,7,
  4. Guillermo Amescua8,
  5. Cyril Zakka9,
  6. Timothy J Archer3,4,
  7. Jeffrey Yammine1,
  8. Elsa Lamah1,
  9. Michèle Haykal10,
  10. Shady T Awwad11
  1. 1Faculty of Medicine, American University of Beirut, Beirut, Lebanon
  2. 2Casey Eye Institute, Pregon Health & Science University, Portland, OR, USA
  3. 3London Vision Clinic, London, UK
  4. 4Reinstein Vision, London, UK
  5. 5Columbia University Medical Center, New York, NY, USA
  6. 6Sorbonne Université, Paris, France
  7. 7Biomedical Science Research Institute, Ulster University, Coleraine, UK
  8. 8Ophthalmology, Bascom Palmer Eye Institute, Miami, Florida, USA
  9. 9Department of Cardiothoracic Surgery, Stanford University, Stanford, California, USA
  10. 10Faculty of Medicine, Saint Joseph University, Beirut, Lebanon
  11. 11Department of Ophthalmology, American University of Beirut Medical Center, Beirut, Lebanon
  1. Correspondence to Dr Shady T Awwad, Department of Ophthalmology, American University of Beirut Medical Center, Beirut, Lebanon; sawwad{at}


Aims To develop a generative adversarial network (GAN) capable of generating realistic high-resolution anterior segment optical coherence tomography (AS-OCT) images.

Methods This study included 142 628 AS-OCT B-scans from the American University of Beirut Medical Center. The Style and WAvelet based GAN architecture was trained to generate realistic AS-OCT images and was evaluated through the Fréchet Inception Distance (FID) Score and a blinded assessment by three refractive surgeons who were asked to distinguish between real and generated images. To assess the suitability of the generated images for machine learning tasks, a convolutional neural network (CNN) was trained using a dataset of real and generated images over a classification task. The generated AS-OCT images were then upsampled using an enhanced super-resolution GAN (ESRGAN) to achieve high resolution.

Results The generated images exhibited visual and quantitative similarity to real AS-OCT images. Quantitative similarity assessed using FID scored an average of 6.32. Surgeons scored 51.7% in identifying real versus generated images which was not significantly better than chance (p value >0.3). The CNN accuracy improved from 78% to 100% when synthetic images were added to the dataset. The ESRGAN upsampled images were objectively more realistic and accurate compared with traditional upsampling techniques by scoring a lower Learned Perceptual Image Patch Similarity of 0.0905 compared with 0.4244 of bicubic interpolation.

Conclusions This study successfully developed and leveraged GANs capable of generating high-definition synthetic AS-OCT images that are realistic and suitable for machine learning and image analysis tasks.

  • Anterior chamber
  • Imaging
  • Phakic Intraocular Lenses
  • Treatment Lasers

Data availability statement

No data are available.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • X @AssafJad

  • Contributors JFA: Conceptualisation, methodology, software, formal analysis, writing—original draft, visualisation. STA: guarantor, supervision, validation, project administration, writing—review and editing. AAM: software, data curation, writing—original draft. DZR: validation, writing—review and editing. GA: validation, writing—review and editing. CZ: methodology, writing—review and editing. TJA: formal analysis, writing—review and editing. JY: data curation, writing—original draft. EL: data curation, writing—original draft. MH: data curation, writing—original draft.

  • Funding This research was partially funded by the Suheil Jamil Muasher Endowed Medical Student Research Award.

  • Competing interests DZR is a consultant for Carl Zeiss Meditec (Carl Zeiss Meditec AG, Jena, Germany). DZR is also a consultant for CSO Italia (Florence, Italy) and has a proprietary interest in the Artemis technology (ArcScan Inc, Golden, Colorado) through patents administered by the Cornell Center for Technology Enterprise and Commercialization (CCTEC), Ithaca, New York. The remaining authors have no proprietary or financial interest in the materials presented herein.

  • Provenance and peer review Not commissioned; externally peer reviewed.