June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Warp speed multi-aperture adaptive optics scanning laser ophthalmoscopy using machine learning
Author Affiliations & Notes
  • Jongwan Park
    Biomedical Engineering, Duke University, Durham, North Carolina, United States
  • Kristen Hagan
    Biomedical Engineering, Duke University, Durham, North Carolina, United States
  • Theodore DuBose
    Biomedical Engineering, Duke University, Durham, North Carolina, United States
  • Ryan P McNabb
    Ophthalmology, Duke University School of Medicine, Durham, North Carolina, United States
  • Alfredo Dubra
    Byers Eye Institute, Stanford University, Stanford, California, United States
  • Joseph Izatt
    Biomedical Engineering, Duke University, Durham, North Carolina, United States
    Ophthalmology, Duke University School of Medicine, Durham, North Carolina, United States
  • Sina Farsiu
    Biomedical Engineering, Duke University, Durham, North Carolina, United States
    Ophthalmology, Duke University School of Medicine, Durham, North Carolina, United States
  • Footnotes
    Commercial Relationships   Jongwan Park Duke University, Code P (Patent); Kristen Hagan Duke University, Code P (Patent); Theodore DuBose Duke University, Code P (Patent); Ryan McNabb Duke University, Code P (Patent); Alfredo Dubra None; Joseph Izatt Duke University, Code P (Patent); Sina Farsiu Duke University, Code P (Patent)
  • Footnotes
    Support  NSF 1902904
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 4332. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Jongwan Park, Kristen Hagan, Theodore DuBose, Ryan P McNabb, Alfredo Dubra, Joseph Izatt, Sina Farsiu; Warp speed multi-aperture adaptive optics scanning laser ophthalmoscopy using machine learning. Invest. Ophthalmol. Vis. Sci. 2023;64(8):4332.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Adaptive optics scanning light ophthalmoscopes (AOSLOs) incorporating non-confocal light detection at multiple offset aperture positions allow for the visualization of various transparent retinal structures. However, the sequential acquisition of images using non-confocal detectors at multiple offsets substantially increases image acquisition time. By leveraging the recent advances in machine learning and optics as an integrated technology, we propose a novel design for multi-aperture AOSLO, called deep-compressed AOSLO (DCAOSLO), which rapidly samples and reconstructs non-confocal light while simultaneously performing confocal imaging.

Methods : Non-confocal light was compressively sampled by a retinal-conjugate digital micromirror device (DMD) by displaying pseudo-random patterns refreshed at the imaging raster line rate. The pseudo-random DMD patterns were designed to sample a circular area 20 Airy discs in diameter (ADD) divided into 12 sub-apertures. The non-confocal sub-aperture images were then reconstructed using an untrained deep generative convolutional neural network. The proposed approach was validated against the conventional offset aperture (OA) detection by imaging the photoreceptor mosaic of 6 healthy adults (Fig. 1).

Results : DCAOSLO sub-aperture images had better quality than the OA images with similar acquisition time and were comparable to averaged OA images captured with ~96 times higher image acquisition time.

Conclusions : Subjective evaluation of sub-aperture images suggests that DCAOSLO is a promising path for reducing non-confocal photoreceptor mosaic imaging time and/or retinal light exposure in healthy subjects. Importantly, this approach can be integrated into existing AOSLOs with minimal hardware modification and facilitate the clinical adoption of non-confocal AOSLO imaging.

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

 

Comparison between offset aperture (OA) imaging and the proposed DCAOSLO imaging methods. Images in column A were generated using 1157 OA frames, where 93-102 frames were averaged per sub-aperture. Images in column B were generated using 12 OA frames, a single frame per sub-aperture. DCAOSLO images of column C were captured ~96 times faster than those in column A, at an acquisition speed similar to the images in Column B. Aperture diagrams represent which sub-aperture images were differentiated to generate images shown in each column. Scale bar 25 µm.

Comparison between offset aperture (OA) imaging and the proposed DCAOSLO imaging methods. Images in column A were generated using 1157 OA frames, where 93-102 frames were averaged per sub-aperture. Images in column B were generated using 12 OA frames, a single frame per sub-aperture. DCAOSLO images of column C were captured ~96 times faster than those in column A, at an acquisition speed similar to the images in Column B. Aperture diagrams represent which sub-aperture images were differentiated to generate images shown in each column. Scale bar 25 µm.

×
×

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.

×