June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Comparison of Deep Learning Glaucoma Detection Using Optic Nerve Head Fundus Photos and Optical Coherence Tomography
Author Affiliations & Notes
  • Mark Christopher
    Hamilton Glaucoma Center, Shiley Eye Institute, Viterbi Family Department of Ophthalmology, University of California San Diego, La Jolla, California, United States
  • Christopher Bowd
    Hamilton Glaucoma Center, Shiley Eye Institute, Viterbi Family Department of Ophthalmology, University of California San Diego, La Jolla, California, United States
  • Evan Walker
    Hamilton Glaucoma Center, Shiley Eye Institute, Viterbi Family Department of Ophthalmology, University of California San Diego, La Jolla, California, United States
  • Akram Belghith
    Hamilton Glaucoma Center, Shiley Eye Institute, Viterbi Family Department of Ophthalmology, University of California San Diego, La Jolla, California, United States
  • Michael Henry Goldbaum
    Hamilton Glaucoma Center, Shiley Eye Institute, Viterbi Family Department of Ophthalmology, University of California San Diego, La Jolla, California, United States
  • Jasmin Rezapour
    Hamilton Glaucoma Center, Shiley Eye Institute, Viterbi Family Department of Ophthalmology, University of California San Diego, La Jolla, California, United States
  • Massimo A. Fazio
    Callahan Eye Hospital, Heersink School of Medicine, University of Alabama - Birmingham, Birmingham, Alabama, United States
  • Christopher A Girkin
    Callahan Eye Hospital, Heersink School of Medicine, University of Alabama - Birmingham, Birmingham, Alabama, United States
  • Gustavo De Moraes
    Bernard and Shirlee Brown Glaucoma Research Laboratory, Edward S. Harkness Eye Institute, Department of Ophthalmology, Columbia University Irving Medical Center, New York, New York, United States
  • Jeffrey M Liebmann
    Bernard and Shirlee Brown Glaucoma Research Laboratory, Edward S. Harkness Eye Institute, Department of Ophthalmology, Columbia University Irving Medical Center, New York, New York, United States
  • Robert N Weinreb
    Hamilton Glaucoma Center, Shiley Eye Institute, Viterbi Family Department of Ophthalmology, University of California San Diego, La Jolla, California, United States
  • Andrzej Grzybowski
    Department of Ophthalmology, Uniwersytet Warminsko-Mazurski w Olsztynie, Olsztyn, Poland
    Institute for Research in Ophthalmology, Poznan, Poland
  • Linda Zangwill
    Hamilton Glaucoma Center, Shiley Eye Institute, Viterbi Family Department of Ophthalmology, University of California San Diego, La Jolla, California, United States
  • Footnotes
    Commercial Relationships   Mark Christopher National Eye Institute, Code F (Financial Support); Christopher Bowd None; Evan Walker None; Akram Belghith None; Michael Goldbaum None; Jasmin Rezapour German Research Foundation, German Ophthalmological Society, Code F (Financial Support); Massimo Fazio National Eye Institute, EyeSight Foundation of Alabama, Research to Prevent Blindness, Heidelberg Engineering, Code F (Financial Support); Christopher Girkin National Eye Institute, EyeSight Foundation of Alabama, Research to Prevent Blindness, Heidelberg Engineering, Code F (Financial Support); Gustavo De Moraes Novartis, Galimedix, Belite, Reichert, Carl Zeiss Meditec, Perfuse Therapeutics, Code C (Consultant/Contractor), Ora Clinical, Code E (Employment), Heidelberg Engineering, Topcon, Code R (Recipient); Jeffrey Liebmann Alcon, Allergan, Bausch & Lomb, Carl Zeiss Meditec, Heidelberg Engineering, Reichert, Valeant Pharmaceuticals, Code C (Consultant/Contractor), Bausch & Lomb, Carl Zeiss Meditec, Heidelberg Engineering, National Eye Institute, Novartis, Optovue, Reichert Technologies, Research to Prevent Blindness, Code F (Financial Support); Robert Weinreb Abbvie, Aerie Pharmaceuticals, Allergan, Equinox, Eyenovia, Nicox, Topcon, Code C (Consultant/Contractor), eidelberg Engineering, Carl Zeiss Meditec, Konan Medical, Optovue, Centervue, Bausch & Lomb, Topcon, National Eye Institute, Research to Prevent Blindness, Code F (Financial Support), Toromedes, Carl Zeiss Meditec, Topcon, Code P (Patent); Andrzej Grzybowski Viatris, Polpharma, Thea, Code C (Consultant/Contractor); Linda Zangwill Abbvie, Digital Diagnostics, Code C (Consultant/Contractor), National Eye Institute, Carl Zeiss Meditec Inc., Heidelberg Engineering GmbH, Optovue Inc., Topcon Medical Systems Inc., Code F (Financial Support), Carl Zeiss Meditec, Code P (Patent)
  • Footnotes
    Support  NEI: K99 EY030942, EY11008, P30 EY022589, EY026590, EY022039, EY021818, EY023704, EY029058, T32 EY026590, R21 EY027945, Genentech, Inc. Unrestricted grant from Research to Prevent Blindness (New York, NY)
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2035 – A0476. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Mark Christopher, Christopher Bowd, Evan Walker, Akram Belghith, Michael Henry Goldbaum, Jasmin Rezapour, Massimo A. Fazio, Christopher A Girkin, Gustavo De Moraes, Jeffrey M Liebmann, Robert N Weinreb, Andrzej Grzybowski, Linda Zangwill; Comparison of Deep Learning Glaucoma Detection Using Optic Nerve Head Fundus Photos and Optical Coherence Tomography. Invest. Ophthalmol. Vis. Sci. 2022;63(7):2035 – A0476.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To compare the accuracy of deep learning (DL) in detecting glaucoma using optic nerve head (ONH) fundus photos versus optical coherence tomography (OCT) across different patient populations, disease severity, and axial lengths.

Methods : Longitudinal imaging and 24-2 visual field (VF) testing was collected on a Diagnostic Innovations in Glaucoma Study (DIGS) and African Descent and Glaucoma Study (ADAGES) cohort of 546 glaucoma (935 eyes) and 346 healthy (626 eyes) patients. The dataset included 20,828 ONH fundus photos and 25,751 unsegmented Spectralis circle scans. Eyes in the glaucoma group had repeatable glaucomatous VF damage or glaucomatous ONH damage based on clinical ONH examination or expert review of photos. Healthy eyes were required to have normal appearing ONHs and VF results. Participants were randomly divided into independent training (85%), validation (5%), and test (10%) sets. Accuracy was evaluated using area under the receiver operating characteristic curve (AUC) and evaluated as a function of participant ancestry (African vs. European), glaucoma severity (24-2 MD > -4.0 dB vs. 24-2 MD <= -4.0 dB), and axial length (AL <= 25 mm vs. AL > 25 mm).

Results : Fundus DL models achieved an AUC of 0.91 (95% CI: 0.89 – 0.92), significantly higher (p=1.3x10-5) than OCT DL models, AUC = 0.86 (95% CI: 0.84 – 0.87). Fundus DL accuracy was significantly higher (p<1x10-6) in African (AUC = 0.97) vs. European (AUC = 0.85) descent participants, significantly higher (p p<1x10-6) in eyes with 24-2 MD <= -4.0 dB (AUC = 0.99) vs. 24-2 MD > -4.0 dB (AUC = 0.88), and significantly higher (p=0.03) in eyes with AL > 25 mm (AUC = 0.94) vs. AL <= 25 mm (AUC = 0.90). ONH OCT DL accuracy was significantly higher (p=0.001) in European (AUC = 0.87) vs. African (AUC = 0.81) descent participants, significantly higher (p<1x10-6) in eyes with 24-2 MD <= -4.0 dB (AUC = 0.95) vs. 24-2 MD > -4.0 dB (AUC = 0.81), and comparable (p=0.82) in eyes with AL > 25 mm (AUC = 0.84) vs. AL <= 25 mm (AUC = 0.85).

Conclusions : Photo-based DL glaucoma detection significantly outperformed OCT-based detection. In both cases, ancestry and disease severity had a significant impact on DL model accuracy. These findings underscore the importance of understanding how patient characteristics, disease severity, and reference standard definitions influence the performance of DL models in glaucoma.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

×
×

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.

×