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
The Impact of Deep Learning Assistance on Ophthalmologists’ Prediction of Recurrence in neovascular AMD
Author Affiliations & Notes
  • Chan Ho Lee
    Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
  • Boa Jang
    Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
  • Young-Gon Kim
    Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
  • Eun Kyoung Lee
    Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
    Seoul National University College of Medicine, Seoul, Korea (the Republic of)
  • Footnotes
    Commercial Relationships   Chan Ho Lee None; Boa Jang None; Young-Gon Kim None; Eun Kyoung Lee None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2024, Vol.65, 2345. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Chan Ho Lee, Boa Jang, Young-Gon Kim, Eun Kyoung Lee; The Impact of Deep Learning Assistance on Ophthalmologists’ Prediction of Recurrence in neovascular AMD. Invest. Ophthalmol. Vis. Sci. 2024;65(7):2345.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Neovascular aged-related macular degeneration (nAMD) is the leading cause of irreversible blindness and has a highly heterogeneous course among patients. In this study, we aimed to predict the first recurrence after the loading phase using optical coherence tomography (OCT) images of nAMD patients, and to investigate the impact of deep learning (DL) on the recurrence prediction performance of a group of experts.

Methods : We reviewed the records of patients diagnosed with treatment-naïve nAMD at a single center. OCT images were obtained at baseline and after the loading phase. The recurrence classification was built using ResNet50. Twenty readers assessed all OCT images using binary-point scale to predict the first recurrence in five sessions. The data provided for five sets of experiments was as follows: OCT at baseline (Set 1), OCT after the loading phase (Set 2), Set 1 + Set 2 images (Set 3), Set 3 + patients’ clinical information (Set 4), Set 4 + heatmap of DL (Set 5). (Figure 1, A-D)

Results : A total of 1,295 eyes of 1,172 patients were included to establish the DL algorithm and a test set of 149 eyes of 130 patients were evaluated. The accuracy of the classification performance using the DL was 0.744, while that of the expert group was 0.562 ± 0.034 (Set 1), 0.665 ± 0.044 (Set 2), 0.663 ± 0.051 (Set 3), 0.649 ± 0.054 (Set 4), and 0.678 ± 0.490 (Set 5) (Figure 2). There was no difference of performance in predicting the first recurrence based on years’ experience or subspecialty from the expert group. Significantly higher accuracy (AUROC; Set 4 vs. Set 5, P < 0.001) and consistency (Fleiss kappa; Set 4 = 0.369, Set 5 = 0.480, all P < 0.005) among different readers were obtained with the assistance of the DL model.

Conclusions : Predicting the first recurrence based on OCT images is still challenging, and this was also a difficult task for the expert group. However, current study evaluated the practical feasibility of a prediction tool using OCT-based DL algorithms in patients with nAMD. An automated prediction system may assist in establishing patient-specific treatment plans and the provision of individualized medical care for patients with nAMD.

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

 

Schematic of the workflow for the reader study.

Schematic of the workflow for the reader study.

 

The AUROC of nineteen ophthalmologists in each reading set.

The AUROC of nineteen ophthalmologists in each reading set.

×
×

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.

×