June 2021
Volume 62, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2021
Combining artificial intelligence and robotics: a semi-automated optical coherence tomography-based approach for posterior eye disease screening
Author Affiliations & Notes
  • Ailin Song
    Duke University School of Medicine, Durham, North Carolina, United States
  • Pablo Ortiz
    Department of Biomedical Engineering, Duke University, Durham, North Carolina, United States
  • Mark Draelos
    Duke University School of Medicine, Durham, North Carolina, United States
    Department of Biomedical Engineering, Duke University, Durham, North Carolina, United States
  • Stefanie G Schuman
    Department of Ophthalmology, Duke University, Durham, North Carolina, United States
  • Glenn J Jaffe
    Department of Ophthalmology, Duke University, Durham, North Carolina, United States
  • Sina Farsiu
    Department of Biomedical Engineering, Duke University, Durham, North Carolina, United States
    Department of Ophthalmology, Duke University, Durham, North Carolina, United States
  • Joseph Izatt
    Department of Biomedical Engineering, Duke University, Durham, North Carolina, United States
    Department of Ophthalmology, Duke University, Durham, North Carolina, United States
  • Ryan P McNabb
    Department of Ophthalmology, Duke University, Durham, North Carolina, United States
  • Anthony N Kuo
    Department of Biomedical Engineering, Duke University, Durham, North Carolina, United States
    Department of Ophthalmology, Duke University, Durham, North Carolina, United States
  • Footnotes
    Commercial Relationships   Ailin Song, None; Pablo Ortiz, None; Mark Draelos, None; Stefanie Schuman, None; Glenn Jaffe, None; Sina Farsiu, None; Joseph Izatt, Kirkland &Ellis LLP (C), Leica Microsystems (P), Leica Microsystems (R), St. Jude Medical (P), St. Jude Medical (R); Ryan McNabb, None; Anthony Kuo, None
  • Footnotes
    Support  Duke CTSA grant UL1TR002553, RPB Medical Student Eye Research Fellowship, NIH R01-EY029302
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 120. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Ailin Song, Pablo Ortiz, Mark Draelos, Stefanie G Schuman, Glenn J Jaffe, Sina Farsiu, Joseph Izatt, Ryan P McNabb, Anthony N Kuo; Combining artificial intelligence and robotics: a semi-automated optical coherence tomography-based approach for posterior eye disease screening. Invest. Ophthalmol. Vis. Sci. 2021;62(8):120.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Despite its importance in systemic diseases such as diabetes, the posterior eye is difficult to examine for non-specialists. To improve eye care in non-ophthalmology settings, we developed a semi-automated approach with potential for posterior eye disease screening, coupling a robotically-aligned optical coherence tomography system and a deep learning (DL) algorithm to classify the images.

Methods : Between August and October 2020, patients seen at the Duke Eye Center and healthy volunteers (age≥18) were imaged with a custom, robotically-aligned OCT (RAOCT) system following clinical eye exam (Fig 1). Using transfer learning, we adapted a preexisting convolutional neural network (Szegedy C, et al. CVPR 2016) to train a DL algorithm to classify OCT images as normal vs. abnormal. The model was trained and validated on two publicly available OCT datasets (Kermany DS, et al. Cell 2018; Srinivasan PP, et al. BOEx 2014) and two of our own RAOCT volumes. For external testing, the top-performing model based on validation was applied to a representative averaged B-scan from each of the remaining RAOCT volumes. The model’s performance was evaluated against the reference standard clinical diagnosis. Saliency maps were created to visualize the areas contributing most to the model predictions.

Results : The training and validation datasets included 87,697 OCT images, of which 59,743 were abnormal. The top-performing DL model had a training accuracy of 96% and a validation accuracy of 99%. For external testing, 43 eyes of 27 subjects were imaged with the RAOCT system. Compared to clinical diagnoses, the model correctly labeled 18 out of 22 normal averaged B-scans and 18 out of 21 abnormal averaged B-scans. In the testing set, the model had an AUC for the detection of pathology of 0.92. For the correctly predicted scans, saliency maps identified the areas contributing most to the DL algorithm’s predictions, which matched the regions of greatest clinical importance (Fig 2).

Conclusions : This is the first study to combine a DL model with a robotic OCT system, demonstrating a potential platform to automate eye disease screening.

This is a 2021 ARVO Annual Meeting abstract.

 

Fig 1. A) Robotically-aligned OCT system. B) Example normal B-scan obtained from the system.

Fig 1. A) Robotically-aligned OCT system. B) Example normal B-scan obtained from the system.

 

Fig 2. Saliency maps highlighting areas of pathology in A) myopic degeneration, B) macular edema, and C) multifocal choroiditis and panuveitis.

Fig 2. Saliency maps highlighting areas of pathology in A) myopic degeneration, B) macular edema, and C) multifocal choroiditis and panuveitis.

×
×

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.

×