June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Explainabilty of a Deep Learning System Able to Accurately Detect Arteritic Anterior Ischemic Optic Neuropathy from Giant Cell Arteritis on Fundus Photographs
Author Affiliations & Notes
  • Dan Milea
    Visual Neuroscience, Singapore Eye Research Institute, Singapore, Singapore
  • Zhiqun Tang
    Visual Neuroscience, Singapore Eye Research Institute, Singapore, Singapore
  • Riccardo Sadun
    Universita Cattolica del Sacro Cuore Medical School, Rome, Italy
  • Marc Dinkin
    Weill Cornell Medicine, New York, New York, United States
  • Wolf Lagreze
    Freiburg University, Freiburg, Germany
  • Kanchalika Sathianvichitr
    Visual Neuroscience, Singapore Eye Research Institute, Singapore, Singapore
  • Cristiano Oliveira
    Weill Cornell Medicine, New York, New York, United States
  • Jing Liang Loo
    Singapore National Eye Centre, Singapore, Singapore, Singapore
  • Reuben Foo
    Singapore National Eye Centre, Singapore, Singapore, Singapore
  • Shweta Singhal
    Singapore National Eye Centre, Singapore, Singapore, Singapore
  • Sharon Tow
    Singapore National Eye Centre, Singapore, Singapore, Singapore
  • Federico Sadun
    Rome Eye Hospital, Italy
  • Neil R Miller
    Johns Hopkins University, Baltimore, Maryland, United States
  • Nancy J Newman
    Emory University, Atlanta, Georgia, United States
  • Valerie Biousse
    Emory University, Atlanta, Georgia, United States
  • Raymond P. Najjar
    Visual Neuroscience, Singapore Eye Research Institute, Singapore, Singapore
    National University of Singapore, Singapore, Singapore
  • Footnotes
    Commercial Relationships   Dan Milea Optomed, FInland, Code C (Consultant/Contractor); Zhiqun Tang None; Riccardo Sadun None; Marc Dinkin None; Wolf Lagreze None; Kanchalika Sathianvichitr None; Cristiano Oliveira None; Jing Liang Loo None; Reuben Foo None; Shweta Singhal None; Sharon Tow None; Federico Sadun None; Neil Miller None; Nancy Newman None; Valerie Biousse None; Raymond Najjar None
  • Footnotes
    Support  Supported by the Singapore National Medical Research Council (Clinician Scientist Individual Research grant CIRG18Nov-0013), the Duke-NUS Medical School, Ophthalmology and Visual Sciences Academic Clinical Program grant (05/FY2019/P2/06-A60).
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 1874. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Dan Milea, Zhiqun Tang, Riccardo Sadun, Marc Dinkin, Wolf Lagreze, Kanchalika Sathianvichitr, Cristiano Oliveira, Jing Liang Loo, Reuben Foo, Shweta Singhal, Sharon Tow, Federico Sadun, Neil R Miller, Nancy J Newman, Valerie Biousse, Raymond P. Najjar; Explainabilty of a Deep Learning System Able to Accurately Detect Arteritic Anterior Ischemic Optic Neuropathy from Giant Cell Arteritis on Fundus Photographs. Invest. Ophthalmol. Vis. Sci. 2023;64(8):1874.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : The urgent and accurate identification of the arteritic form of anterior ischemic optic neuropathy (AAION) from giant cell arteritis is of critical therapeutic importance. Its distinction from the more common form of non arteritic ischemic optic neuropathy (NAION), not requiring an urgent intervention is difficult. The aim of this study was to evaluate the performance and the explainability of a deep learning system (DLS) to distinguish AAION from NAION on standard ocular fundus photographs.

Methods : This multicenter, multiethnic, retrospective study (26 neuro-ophthalmology centers in 18 countries) included 825 patients (852 pathologic optic disc photographs, including 138 AAION and 714 NAION at the acute stage) for the development, training and validation of the DLS. External testing was performed on 121 patients (137 images) collected from four independent expert centers in the USA and in Europe, including 46 AAION images and 91 NAION images. Individual heatmaps were pooled to obtain a global activation display for each condition.

Results : The DLS showed an excellent performance for the discrimination of AAION from NAION [Area under curve (AUC): 0.98 (0.96-1.0), sensitivity: 97.8% (95.8%-100.0%), specificity: 93.4% (89.5%-98.0%)]. The DLS displayed explainable, distinct patterns of pooled class activation maps in the optic disc regions (i.e., diffuse area of interest in AAION compared to a specific, well delineated inferior optic disc area in NAION).

Conclusions : A trained DLS can accurately classify acute AAION and NAION on ocular fundus photographs alone, without any clinical or biomarker information. The DLS displays distinct, explainable optic disc class activation patterns, compatible with the pathophysiology of each condition. With further prospective validation, this DLS could become an automated diagnostic aid to help discriminate AAION from NAION in acute clinical settings.

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

 

Distinct pooled activation heatmaps generated by the DLS in each group of patients, allowing for the accurate discrimination of AAION vs NAION. The pooled heatmaps provide explainable features, compatible with the pathophysiology of each condition, i.e. diffuse optic disc ischemia in AAION versus a restricted inferior area of interest in NAION.

Distinct pooled activation heatmaps generated by the DLS in each group of patients, allowing for the accurate discrimination of AAION vs NAION. The pooled heatmaps provide explainable features, compatible with the pathophysiology of each condition, i.e. diffuse optic disc ischemia in AAION versus a restricted inferior area of interest in NAION.

×
×

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.

×