July 2019
Volume 60, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2019
A Deep Learning Algorithm to Quantify Neuroretinal Rim Loss from Optic Disc Photographs
Author Affiliations & Notes
  • Atalie C. Thompson
    Duke Eye Center, Durham, North Carolina, United States
  • Alessandro A Jammal
    Duke Eye Center, Durham, North Carolina, United States
  • Felipe A Medeiros
    Duke Eye Center, Durham, North Carolina, United States
  • Footnotes
    Commercial Relationships   Atalie Thompson, None; Alessandro Jammal, None; Felipe Medeiros, Allergan (C), Allergan (F), Bausch & Lomb (F), Carl Zeiss Meditec (C), Carl Zeiss Meditec (F), Heidelberg Engineering (F), Merck (F), nGoggle, Inc. (F), Novartis (C), Reichert (C), Reichert (R), Sensimed (C)
  • Footnotes
    Support  Supported in part by National Institutes of Health/National Eye Institute grant EY027651 (FAM), EY025056 (FAM) and EY021818 (FAM).
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 5577. doi:https://doi.org/
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Atalie C. Thompson, Alessandro A Jammal, Felipe A Medeiros; A Deep Learning Algorithm to Quantify Neuroretinal Rim Loss from Optic Disc Photographs. Invest. Ophthalmol. Vis. Sci. 2019;60(9):5577. doi: https://doi.org/.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To train and test a deep learning (DL) algorithm that quantifies glaucomatous neuroretinal damage on fundus photographs using the minimum rim width relative to Bruch’s membrane opening (BMO-MRW) from spectral domain-optical coherence tomography (SDOCT).

Methods : 9,282 pairs of optic disc photographs and SDOCT optic nerve head scans from 927 eyes of 490 subjects were randomly divided into the training plus validation (80%) and test sets (20%). A DL convolutional neural network was trained to predict the SDOCT BMO-MRW global and sector values when evaluating optic disc photographs. The predictions of the DL network were compared to the actual SDOCT measurements. The area under the receiver operating curve (AUC) was used to evaluate the ability of the network to discriminate eyes with glaucomatous visual field loss from normal eyes. Gradient-weighted class activation maps (Figure 1) were built over the input images to indicate the importance of each location of the image to the class under consideration.

Results : The DL predictions of global BMO-MRW from all optic disc photos in the test set (mean±standard deviation [SD]: 228.8±63.1mm) were highly correlated with the observed values from SDOCT (mean±SD: 226.0±73.8mm) (Pearson’s r=0.88; R2=77%; P<0.001; Figure 2), with mean absolute error of the predictions of 27.8mm. The AUCs for discriminating glaucomatous from healthy eyes with the DL predictions and actual SDOCT global BMO-MRW measurements were 0.945 (95% CI:0.874-0.980) and 0.933 (95% CI:0.856-0.975), respectively (P=0.587).

Conclusions : A DL network can be trained to quantify the amount of neuroretinal damage on optic disc photographs using SDOCT BMO-MRW as a reference. This algorithm showed high accuracy for glaucoma detection, and may overcome limitations of algorithms trained by human gradings of disc photos.

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

 

Figure 1. Activation heatmaps showing the areas of the optic disc photograph that were most important for the deep learning algorithm predictions in an example of a healthy (A) and glaucomatous (B) eye

Figure 1. Activation heatmaps showing the areas of the optic disc photograph that were most important for the deep learning algorithm predictions in an example of a healthy (A) and glaucomatous (B) eye

 

Figure 2. Scatterplot illustrating the relationship between predictions obtained by the deep learning algorithm evaluating optic disc photographs and actual global minimum rim width relative to Bruch’s membrane opening (BMO-MRW) thickness measurements from spectral domain-optical coherence tomography (SDOCT). Data is from the independent test set.

Figure 2. Scatterplot illustrating the relationship between predictions obtained by the deep learning algorithm evaluating optic disc photographs and actual global minimum rim width relative to Bruch’s membrane opening (BMO-MRW) thickness measurements from spectral domain-optical coherence tomography (SDOCT). Data is from the independent test 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.

×