June 2020
Volume 61, Issue 7
Free
ARVO Annual Meeting Abstract  |   June 2020
Automatic cellular level differentiation of glaucomatous and healthy eyes via deep learning-based adaptive optics OCT analysis
Author Affiliations & Notes
  • Somayyeh Soltanian-Zadeh
    Biomedical Engineering, Duke University, Durham, North Carolina, United States
  • Zhuolin Liu
    U.S. Food and Drug Administration, Silver Spring, Maryland, United States
  • Ricardo Villanueva
    University of Maryland School of Medicine, Baltimore, Maryland, United States
  • Osamah Saeedi
    University of Maryland School of Medicine, Baltimore, Maryland, United States
  • Kazuhiro Kurokawa
    School of Optometry, Indiana University, Bloomington, Indiana, United States
  • Hae Won Jung
    School of Optometry, Indiana University, Bloomington, Indiana, United States
  • Donald Thomas Miller
    School of Optometry, Indiana University, Bloomington, Indiana, United States
  • Daniel Hammer
    U.S. Food and Drug Administration, Silver Spring, Maryland, United States
  • Sina Farsiu
    Biomedical Engineering, Duke University, Durham, North Carolina, United States
    Opthalmology, Duke University Eye Center, Durham, North Carolina, United States
  • Footnotes
    Commercial Relationships   Somayyeh Soltanian-Zadeh, None; Zhuolin Liu, Indiana University (P); Ricardo Villanueva, None; Osamah Saeedi, Heidelberg Engineering (F), Vasoptic Inc. (F); Kazuhiro Kurokawa, Indiana University (P); Hae Won Jung, None; Donald Miller, Indiana University (P); Daniel Hammer, None; Sina Farsiu, Google (R)
  • Footnotes
    Support  Google Faculty Research Award, National Institutes of Health (R01 EY029808), National Eye Institute (K23EY025014)
Investigative Ophthalmology & Visual Science June 2020, Vol.61, 877. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Somayyeh Soltanian-Zadeh, Zhuolin Liu, Ricardo Villanueva, Osamah Saeedi, Kazuhiro Kurokawa, Hae Won Jung, Donald Thomas Miller, Daniel Hammer, Sina Farsiu; Automatic cellular level differentiation of glaucomatous and healthy eyes via deep learning-based adaptive optics OCT analysis. Invest. Ophthalmol. Vis. Sci. 2020;61(7):877.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Automatic and accurate diagnosis and prognosis of glaucoma at a cellular level.

Methods : A multimodal adaptive optics OCT (AO-OCT) imager [Liu et al., Biomed. Opt. Express, 9(9), 2018] was used to acquire volumes at 12° temporal to the fovea from five glaucoma and four healthy subjects. We developed a weakly supervised deep learning-based method to detect and segment individual retinal ganglion and displaced amacrine cells (RGDCs) from AO-OCT volumes. The algorithm was trained and tested using leave-one-subject-out cross-validation. We calculated cell densities using cell counts and accounting for large blood vessels. We diagnosed individuals through the k-nearest neighbor classifier and compared the results to clinical diagnosis. We diagnosed glaucoma in a subject if the majority of their volumes was classified as glaucoma.

Results : Figure 1a qualitatively shows our results on example volumes compared to manual marking. Our estimated cell densities, obtained in 1.5-2.4 minutes/volume, were similar to the manually determined values, calculated in 0.5-2 hours/volume, for both groups (Figure 1b). Our automatic analysis reflected RGDC diameters of 14.5±2.3 μm and 17.8±1.5 μm, calculated in 2.1-14.3 seconds/volume, for the healthy and glaucoma subjects, respectively, which are in line with previous measurements [Liu et al., Invest. Ophthal. Vis. Sci., 60(9), 2019]. Using our automatic cell densities and mean diameters as features, diagnosis differed in only one healthy subject (Figure 1c), yielding sensitivity = 1 and specificity = 0.75.

Conclusions : Results indicate that we can accurately and rapidly differentiate glaucoma patients from healthy individuals at the cellular-level via our automatic AO-OCT analysis method.

This is a 2020 ARVO Annual Meeting abstract.

 

Figure 1: (a) Left: Markers denote cell centers projected onto the en face plane. Cyan: true positive, red: false negative, yellow: false positive. Right: Overlay of segmentation masks on the en face plane. Different colors denote different cells. Scale bars: 50 μm. (b) Automatically estimated cell densities agree with manual cell densities for both groups. Crosses: individual data points, circles: mean value. (c) Automatic cell densities and average diameters for all volumes. In both groups, each subject is shown with a different marker shape. Circled data denote diagnoses that differed with clinical assessment.

Figure 1: (a) Left: Markers denote cell centers projected onto the en face plane. Cyan: true positive, red: false negative, yellow: false positive. Right: Overlay of segmentation masks on the en face plane. Different colors denote different cells. Scale bars: 50 μm. (b) Automatically estimated cell densities agree with manual cell densities for both groups. Crosses: individual data points, circles: mean value. (c) Automatic cell densities and average diameters for all volumes. In both groups, each subject is shown with a different marker shape. Circled data denote diagnoses that differed with clinical assessment.

×
×

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.

×