June 2020
Volume 61, Issue 7
Free
ARVO Annual Meeting Abstract  |   June 2020
Convolutional neural networks can predict retinal differentiation in stem-cell derived organoids
Author Affiliations & Notes
  • Evgenii Kegeles
    Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, Massachusetts, United States
    Moscow Institute of Physics and Technology, Dolgoprudniy, Russian Federation
  • Anton Naumov
    Ivannikov Institute for System Programming of the Russian Academy of Sciences, Moscow, Russian Federation
  • Evgeny Karpulevich
    Ivannikov Institute for System Programming of the Russian Academy of Sciences, Moscow, Russian Federation
    Moscow Institute of Physics and Technology, Dolgoprudniy, Russian Federation
  • Pavel Volchkov
    Moscow Institute of Physics and Technology, Dolgoprudniy, Russian Federation
  • Petr Y Baranov
    Schepens Eye Research Institute of Massachusetts Eye and Ear, Harvard Medical School, Boston, Massachusetts, United States
  • Footnotes
    Commercial Relationships   Evgenii Kegeles, None; Anton Naumov, None; Evgeny Karpulevich, None; Pavel Volchkov, None; Petr Baranov, None
  • Footnotes
    Support  Research to Prevent Blindness foundation, Massachusetts Lions Eye Research Fund, Russian Academic Excellence project
Investigative Ophthalmology & Visual Science June 2020, Vol.61, 5201. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Evgenii Kegeles, Anton Naumov, Evgeny Karpulevich, Pavel Volchkov, Petr Y Baranov; Convolutional neural networks can predict retinal differentiation in stem-cell derived organoids. Invest. Ophthalmol. Vis. Sci. 2020;61(7):5201.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Retinal tissue differentiation from stem cells using three-dimensional organoids was demonstrated with high efficiency by the group of Dr. Yoshiki Sasai in 2011. This technology has proven useful in drug discovery, disease modelling and cell production for transplantation. However, the process of retinal differentiation is stochastic in nature which results in high variability between different organoids. Current strategy to select organoids with high content of retinal neurons rely on fluorescent reporters. It is highly sensitive and accurate method, though it has limited applicability in GMP cell manufacture for transplantation. Here we attempted to develop an algorithm which can predict the performance of each organoid based on the bright field (BF) image at the early stages of differentiation.

Methods : RxGFP mouse embryonic reporter stem cells (mES)(RIKEN) have been used for all of the differentiation experiments. Cells have been aggregated on day 1 (d1) in U-well transparent 96 well plates (3 000 cells/well). BF images of organoids have been taken on d6 of the differentiation using EVOS Fl Auto microscope. The assessment of retinal differentiation was based on the expression of RxGFP reporter at day 9. To predict early retinal differentiation, we utilized transfer learning approach: ResNet50 and VGG19 convolutional neural networks (CNN) had been pre-trained on ImageNet dataset and then we trained them using 3 000 BF images of our retinal organoids. Labels (retina, non-retina) for all of these organoids have been assigned by two independent experts based on fluorescent images on d9.

Results : The best prediction quality has shown a classifier based on VGG19 architecture which performed with prediction accuracy of 77%. Comparison of human-based classifier with CNN showed that CNN algorithm performs better than the operator: 74% vs 77% of correct predictions respectively.

Conclusions : We demonstrated that retinal differentiation in organoids can be predicted by computer algorithm based on the automated brightfield imaging. This non-subjective approach has significant translational potential as it can be implemented as a selection strategy or a quality control in GMP cell manufacture process for transplantation or used in drug discovery and disease modelling to decrease the variability and simplify data interpretation.

This is a 2020 ARVO Annual Meeting 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.

×