Investigative Ophthalmology & Visual Science Cover Image for Volume 62, Issue 8
June 2021
Volume 62, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2021
Neural network aided erythrocyte stasis characterization based on erythrocyte mediated angiography (EMA)
Author Affiliations & Notes
  • Dongyi Wang
    University of Maryland at College Park, College Park, Maryland, United States
  • Jessica Pottenburgh
    University of Maryland School of Medicine, Baltimore, Maryland, United States
  • Wei Chen Lai
    University of Maryland School of Medicine, Baltimore, Maryland, United States
  • Caroline Simon
    University of Maryland School of Medicine, Baltimore, Maryland, United States
  • Yang Tao
    University of Maryland at College Park, College Park, Maryland, United States
  • Osamah Saeedi
    University of Maryland School of Medicine, Baltimore, Maryland, United States
  • Footnotes
    Commercial Relationships   Dongyi Wang, None; Jessica Pottenburgh, None; Wei Chen Lai, None; Caroline Simon, None; Yang Tao, None; Osamah Saeedi, None
  • Footnotes
    Support  K23 EY025014
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 2110. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Dongyi Wang, Jessica Pottenburgh, Wei Chen Lai, Caroline Simon, Yang Tao, Osamah Saeedi; Neural network aided erythrocyte stasis characterization based on erythrocyte mediated angiography (EMA). Invest. Ophthalmol. Vis. Sci. 2021;62(8):2110.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : EMA is a novel imaging technique permitting direct erythrocytes visualization. With EMA, erythrocytes can be observed in motion or in stasis. One cause of the transient stasis is hypothesized to be a manifestation of status of vasomotion. Erythrocytes in stasis could serve for characterizing hemodynamics status of retinal/choroidal vasculature in systemic and ocular diseases including primary open-angle glaucoma (POAG). This research has two main goals 1) developing a network model to quantify paused erythrocyte densities in EMA; 2) studying differences of relative paused erythrocyte densities for the control and POAG/POAG suspects.

Methods : 1) To train a network, we labelled 3752 cells in 24 temporally averaged (TA) images from 6 eyes (4 POAG/POAG suspects, 2 controls) which can visualize erythrocytes in stasis. Two trained researchers labelled pausing erythrocytes, and consensual labellings were used for network training. A regressional network was designed and trained in the manner of leave-one-eye-out cross validation to count pausing erythrocytes. 2) The trained model was applied to 22 eyes (16 POAG/POAG suspects, 6 controls) from 15 subjects. For each eye, EMA imaged both its peripapillary and macula area with duration of 5-15 seconds. The detected paused erythrocyte densities are 106.6+/-159.27 cells/TA image and 191.25+/-98.08/TA images in the peripapillary and macula area. We defined peripapillary to macular ratio (PMR) as the paused cell count ratio between peripapillary and macula area. Generalized estimating equation (GEE) was used to compare PMR between controls and POAG/POAG suspects.

Results : 1) The cross validated f1-score of trained network is 0.938, which showed no significant difference compared to two trained researchers (p<0.05). 2) The PMR value was 0.94+/-0.23 in controls and 0.49+/-0.18 in POAG/POAG suspects. POAG is statistically significant related to the PMR value(p<0.001).

Conclusions : 1) Regression based neural network is a reliable model to quantify paused erythrocytes in EMA videos. 2) Our results showed differing PMR values between control and POAG/POAG suspects. This new finding manifests PMR can be potentially considered as a pathophysiological biomarker of early stage glaucomatous damage. Further studies are needed to determine the axial location of paused erythrocytes to better explain the phenomenon.

This is a 2021 ARVO Annual Meeting abstract.

 

Paused cell detection results from the network model

Paused cell detection results from the network model

×
×

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.

×