June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Deep learning-based fully automated dry eye disease severity grading system
Author Affiliations & Notes
  • Chang Ho Yoon
    Ophthalmology, Seoul National University College of Medicine, Seoul, Korea (the Republic of)
    Laboratory of Ocular Regenerative Medicine and Immunology, Biomedical Research Institute, Seoul National University Hospital, Seoul, Korea (the Republic of)
  • Seonghwan Kim
    Ophthalmology, Seoul National University College of Medicine, Seoul, Korea (the Republic of)
    Laboratory of Ocular Regenerative Medicine and Immunology, Biomedical Research Institute, Seoul National University Hospital, Seoul, Korea (the Republic of)
  • Daseul Park
    Transdisciplinary Department of Medicine & Advanced Technology, Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
    Interdisciplinary Program in Bioengineering, Graduate School, Seoul National University, Gwanak-gu, Seoul, Korea (the Republic of)
  • Youmin Shin
    Transdisciplinary Department of Medicine & Advanced Technology, Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
    Interdisciplinary Program in Bioengineering, Graduate School, Seoul National University, Gwanak-gu, Seoul, Korea (the Republic of)
  • Mee Kum Kim
    Ophthalmology, Seoul National University College of Medicine, Seoul, Korea (the Republic of)
    Laboratory of Ocular Regenerative Medicine and Immunology, Biomedical Research Institute, Seoul National University Hospital, Seoul, Korea (the Republic of)
  • Hyun Sun Jeon
    Ophthalmology, Seoul National University College of Medicine, Seoul, Korea (the Republic of)
    Ophthalmology, Seoul National University Bundang Hospital, Seongnam, Korea (the Republic of)
  • Young-Gon Kim
    Transdisciplinary Department of Medicine & Advanced Technology, Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
  • Footnotes
    Commercial Relationships   Chang Ho Yoon None; Seonghwan Kim None; Daseul Park None; Youmin Shin None; Mee Kum Kim None; Hyun Sun Jeon None; Young-Gon Kim None
  • Footnotes
    Support  Supported by the Seoul National University Hospital Research Fund (03-2022-2040), the Electronics and Telecommunications Research Institute (ETRI) grant funded by the Korean government (22ZR1100), and a research fund donated by Hyun Hee Kim.
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 1092. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Chang Ho Yoon, Seonghwan Kim, Daseul Park, Youmin Shin, Mee Kum Kim, Hyun Sun Jeon, Young-Gon Kim; Deep learning-based fully automated dry eye disease severity grading system. Invest. Ophthalmol. Vis. Sci. 2023;64(8):1092.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Although the National Eye Institute (NEI) scale for grading corneal fluorescein staining (CFS) is widely used to assess the severity of dry eye disease, it is associated with disadvantages such as low reproducibility and inter-observer variability. This study aimed to develop a clinically applicable fully automated deep learning-based dry eye severity assessment system.

Methods : A total of 1,300 CFS images were obtained from Seoul National University Hospital. The corneal area in the images was divided into five zones and graded by three experts using the NEI scale (range 0-15), and the median value was used as ground truth. The assessment system development consisted of the following three steps: (1) using 1100 CFS images, the U-Net model was trained to segment the corneal region, (2) the classification model was trained using the ImageNet pre-trained VGG16 model with 192x192 pixel-sized 24,559 patches from 200 images, in which regions containing the punctate dots were labeled, and (3) punctate dots were automatically detected using blob detection algorithms by extracting circular regions of green channel images. A punctate dot density map was generated using a sliding window of 32x32 pixels with 50% overlap. Finally, based on the density map, NEI grades were calculated after adding maximum values of five zones. To validate the performance of the proposed model, Pearson’s correlation analysis was conducted between the summation values and ground-truth data. External validation was conducted using 94 images obtained from Seoul National University Bundang Hospital.

Results : The dice coefficient of the segmentation model was 0.96. The performance of the extraction model showed dice coefficients of 0.92 and 0.90 for accuracy, 0.93 and 0.82 for sensitivity, and 0.92 and 0.96 for specificity with a threshold of 0.5 and 0.99, respectively. The Pearson correlation coefficients between the values derived from the deep learning-based system and the ground truth data were 0.9083 (p<0.001) and 0.8727 (p<0.001) in the internal and external validation datasets, respectively.

Conclusions : The fully automated deep learning-based dry eye severity grading system was able to evaluate the CFS score with high accuracy and is expected to be clinically applied.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

×
×

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.

×