July 2020
Volume 61, Issue 9
Free
ARVO Imaging in the Eye Conference Abstract  |   July 2020
Performance validation of B-scan of interest algorithm on normative dataset
Author Affiliations & Notes
  • Sophia Yu
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Hugang Ren
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Niranchana Manivannan
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Gary Lee
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Patricia Sha
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Alline Melo
    Center for Ophthalmic Bioinformatics, Cole Eye Institute, Cleveland, Ohio, United States
  • Thais Conti
    Center for Ophthalmic Bioinformatics, Cole Eye Institute, Cleveland, Ohio, United States
  • Tyler Greenlee
    Center for Ophthalmic Bioinformatics, Cole Eye Institute, Cleveland, Ohio, United States
  • Eric R. Chen
    Case Western Reserve University School of Medicine, Cleveland, Ohio, United States
  • Katherine Talcott
    Center for Ophthalmic Bioinformatics, Cole Eye Institute, Cleveland, Ohio, United States
  • Rishi P. Singh
    Center for Ophthalmic Bioinformatics, Cole Eye Institute, Cleveland, Ohio, United States
  • Neil D'Souza
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Mary Durbin
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Footnotes
    Commercial Relationships   Sophia Yu, Carl Zeiss Meditec, Inc. (E); Hugang Ren, Carl Zeiss Meditec, Inc. (E); Niranchana Manivannan, Carl Zeiss Meditec, Inc. (E); Gary Lee, Carl Zeiss Meditec, Inc. (E); Patricia Sha, Carl Zeiss Meditec, Inc. (E); Alline Melo, Carl Zeiss Meditec, Inc. (F); Thais Conti, Carl Zeiss Meditec, Inc. (F); Tyler Greenlee, Carl Zeiss Meditec, Inc. (F); Eric Chen, Carl Zeiss Meditec, Inc. (F); Katherine Talcott, Carl Zeiss Meditec, Inc. (F); Rishi Singh, Alcon (C), Apellis (F), Carl Zeiss Meditec, Inc. (C), Genentech (C), Graybug (F), Novartis (C), Regeneron (C); Neil D'Souza, Carl Zeiss Meditec, Inc. (E); Mary Durbin, Carl Zeiss Meditec, Inc. (E)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science July 2020, Vol.61, PB0085. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Sophia Yu, Hugang Ren, Niranchana Manivannan, Gary Lee, Patricia Sha, Alline Melo, Thais Conti, Tyler Greenlee, Eric R. Chen, Katherine Talcott, Rishi P. Singh, Neil D'Souza, Mary Durbin; Performance validation of B-scan of interest algorithm on normative dataset. Invest. Ophthalmol. Vis. Sci. 2020;61(9):PB0085.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Optical coherence tomography (OCT) B-scan of interest is a deep learning algorithm that aims to improve workflow efficiency by instantly indicating B-scans that may be of particular interest to a clinician during OCT data review. To set baseline expectations on the performance of this algorithm, it would be useful to look at the result of implementing the algorithm on an external dataset comprised of only normal eyes. In this study, we evaluated the prevalence of B-scans of interest in the Macula reference database (RDB).

Methods : The CIRRUS HD-OCT Macula RDB contains macular thickness data from 282 subjects with normal eyes (282 eyes with one Macular Cube 512x128 scan each) from 7 clinical sites, representative of healthy individuals (ages 19-84) with no history of eye disease. All B-scans of all cubes in the RDB were considered normal. To train the B-scan algorithm, 2 retina specialists annotated 76,544 B-scan images acquired from 598 subjects using CIRRUS™ HD-OCT 5000 (ZEISS, Dublin, CA) based on 8 different pathologies (Tbl 1) or ungradable. A B-scan is considered “of interest” if it has one of the 8 pathology labels graded by at least 1 retina specialist. 148 B-scans were ungradable and excluded. The other 76,396 B-scans were split into training and development sets with 80/20 ratio for model development. A ResNet-50 deep neural network was trained using 61,058 B-scans from the training set and fine-tuned using 15,338 B-scans from the development set. We used a MATLAB-based workflow to generate the inference result.

Results : The algorithm identified 204 B-scans (0.6% of total) as “of interest” and 35,889 B-scans (99.4% of total) as normal. 43 macular cubes (15.3% of the total cubes) were identified as having at least 1 B-scan of interest and good image quality. Of these 43 cubes, we plotted B-scans of interest per cube to the subjects’ age and found no correlation (Fig 1). 3 cubes were found with significantly more B-scans of interest (14, 55, and 49 B-scans) than the rest of the cohort. When checking the clinical notes of the RDB, these 3 cubes were known to contain either epiretinal membrane, a thin retina, or a wavy retina from a recent posterior vitreous detachment.

Conclusions : In this study, we demonstrated that minimal B-scans of interest were identified in a normative dataset. We can expect the algorithm to perform on average with reasonable specificity on macular cube scans of normal eyes.

This is a 2020 Imaging in the Eye Conference abstract.

 

 

×
×

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.

×