Investigative Ophthalmology & Visual Science Cover Image for Volume 65, Issue 7
June 2024
Volume 65, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2024
Validation of deep learning-based automated segmentation of OCT images for choroidal thickness
Author Affiliations & Notes
  • Raman Prasad Sah
    College of Optometry, University of Houston, Houston, Texas, United States
  • Nimesh Bhikhu Patel
    College of Optometry, University of Houston, Houston, Texas, United States
  • Hope M Queener
    College of Optometry, University of Houston, Houston, Texas, United States
  • Lisa A Ostrin
    College of Optometry, University of Houston, Houston, Texas, United States
  • Footnotes
    Commercial Relationships   Raman Prasad Sah None; Nimesh Patel None; Hope Queener None; Lisa Ostrin Vyluma, Code C (Consultant/Contractor), Zeiss, Code C (Consultant/Contractor), Topcon, Code F (Financial Support), Meta, Code F (Financial Support), US 11375890 B2, Code P (Patent)
  • Footnotes
    Support  NEI P30EY007551, NEI R01EY030193
Investigative Ophthalmology & Visual Science June 2024, Vol.65, 6626. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Raman Prasad Sah, Nimesh Bhikhu Patel, Hope M Queener, Lisa A Ostrin; Validation of deep learning-based automated segmentation of OCT images for choroidal thickness. Invest. Ophthalmol. Vis. Sci. 2024;65(7):6626.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Myopia prevalence is increasing worldwide, prompting extensive research efforts to understand underlying mechanisms and develop tools to control its progression. Studies suggest that the choroid, with its dynamic changes in thickness in response to various environmental cues, could play a role in regulating eye growth. However, estimation of choroidal thickness from optical coherence tomography (OCT) images is challenging, and a need exists for an automated tool capable of accurately estimating thickness. This study aimed to validate a neural network-based program designed to automate choroidal segmentation of 3-dimensional OCT scans.

Methods : An automated segmentation algorithm was trained on a Deeplabv3+ network, based on ResNet50 (accuracy 99.25% and loss 0.0229), using a training set of 10,798 manually segmented OCT scans (Spectralis, Heidelberg Engineering). Then, 116 unique OCT scans from 6 different studies were randomly selected and segmented with both the automated and traditional manual programs written in MATLAB. For manual segmentation, the program delineates Bruch’s membrane and requires the user to delineate the choroid-sclera border. For automated segmentation, the network delineates both borders. For both approaches, the average sub-foveal choroidal thickness within a 1-mm diameter was obtained. Bland-Altman analyses and intraclass correlation coefficient (ICC) were used to compare the two estimated choroidal thicknesses.

Results : The overall mean ± SD choroidal thicknesses obtained with automated and manual segmentation were 337.44 ± 79.37 µm and 334.52 ± 89.47 µm, respectively. Bland-Altman analysis indicated a mean ± SD difference of 2.92 ± 30.62 µm, (95% CI -2.71 to 8.55) between methods, and ICC demonstrated excellent agreement (ICC = 0.93, 95% CI 0.91 to 0.95, p<0.05).

Conclusions : Deep learning-based automated segmentation of choroidal thickness showed excellent agreement with manual segmentation. Observed differences between methods were generally small. Being objective and significantly faster than manual segmentation, this automated approach has distinct advantages for estimating choroidal thickness.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

×
×

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.

×