July 2019
Volume 60, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2019
Development of a deep learning image system for detecting referable retinopathy of prematurity
Author Affiliations & Notes
  • Mingzhi Zhang
    Ophthalmology, Joint Shantou International Eye Center, Shantou, China
  • Guihua Zhang
    Ophthalmology, Joint Shantou International Eye Center, Shantou, China
  • Jie Ji
    Shantou University, China
  • ji wang
    Ophthalmology, Joint Shantou International Eye Center, Shantou, China
  • Jianwei Lin
    Ophthalmology, Joint Shantou International Eye Center, Shantou, China
  • Footnotes
    Commercial Relationships   Mingzhi Zhang, None; Guihua Zhang, None; Jie Ji, None; ji wang, None; Jianwei Lin, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 1458. doi:https://doi.org/
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Mingzhi Zhang, Guihua Zhang, Jie Ji, ji wang, Jianwei Lin; Development of a deep learning image system for detecting referable retinopathy of prematurity. Invest. Ophthalmol. Vis. Sci. 2019;60(9):1458. doi: https://doi.org/.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To describe the development and validation of an artificial intelligence-based, deep learning algorithm for the detection of referable retinopathy of prematurity (ROP).

Methods : Based on retinal images by using deep convolutional neural networks, a DLA which can generate detailed class activation maps was developed for ROP screening. The DLA was trained using a data set of 2748 retinal photographs. Each image was previously assigned a reference standard diagnosis based on consensus of image grading and clinical diagnosis by 2 experts (ie, referable ROP screening and non-referable ROP screening). The algorithm was evaluated by 5-fold cross-validation and tested on an independent set of 1000 images. A modified class activaton maps technique was used to interpret the results generated by the DLA.

Results : Compared to the reference standard diagnosis (RSD) by experts, the sensitivity and the specificity for the validation set were 92.9% and t 91.3%, respectively. After being fully trained and validated, all ConvNet models were deployed into production platform, which can be accessed through the Internet.

Conclusions : This artificial intelligence-based DLA can be used with high accuracy in the detection of referable ROP in retinal images. This technology offers potential to increase the efficiency and accessibility of ROP screening programs.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.

×
×

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.

×