A Deep Learning Approach for Accurate Discrimination Between Optic Disc Drusen and Papilledema on Fundus Photographs

Authors:
  • KSKanchalika Md Sathianvichitr
    Kanchalika Md Sathianvichitr
  • RNRaymond P. Najjar
    Raymond P. Najjar
  • TZTang Zhiqun
    Tang Zhiqun
  • JFJ. Alexander Fraser
    J. Alexander Fraser
  • CYChristine W. L. Mbbs(S'Pore) Yau
    Christine W. L. Mbbs(S'Pore) Yau
  • MGMichael J. A. Girard
    Michael J. A. Girard
  • FCFiona Costello
    Fiona Costello
  • MLMung Y. Lin
    Mung Y. Lin
  • WLWolf A. Lagrèze
    Wolf A. Lagrèze
  • CVCatherine Vignal-Clermont
    Catherine Vignal-Clermont
  • CFClare L. Mbbs Fraser
    Clare L. Mbbs Fraser
  • SHSteffen Hamann
    Steffen Hamann
Publisher website
Google Scholar
Abstract
Background: Optic disc drusen (ODD) represent an important differential diagnosis of papilledema caused by intracranial hypertension, but their distinction may be difficult in clinical practice. The aim of this study was to train, validate, and test a dedicated deep learning system (DLS) for binary classification of ODD vs papilledema (including various subgroups within each category), on conventional mydriatic digital ocular fundus photographs collected in a large international multiethnic population. Methods: This retrospective study included 4,508 color fundus images in 2,180 patients from 30 neuro-ophthalmology centers (19 countries) participating in the Brain and Optic Nerve Study with Artificial Intelligence (BONSAI) Group. For training and internal validation, we used 857 ODD images and 3,230 papilledema images, in 1,959 patients. External testing was performed on an independent data set (221 patients), including 207 images with ODD (96 visible and 111 buried), provided by 3 centers of the Optic Disc Drusen Studies Consortium, and 214 images of papilledema (92 mild-to-moderate and 122 severe) from a previously validated study. Results: The DLS could accurately distinguish between all ODD and papilledema (all severities included): area under the receiver operating characteristic curve (AUC) 0.97 (95% confidence interval [CI], 0.96–0.98), accuracy 90.5% (95% CI, 88.0%–92.9%), sensitivity 86.0% (95% CI, 82.1%–90.1%), and specificity 94.9% (95% CI, 92.3%–97.6%). The performance of the DLS remained high for discrimination of buried ODD from mild-to-moderate papilledema: AUC 0.93 (95% CI, 0.90–0.96), accuracy 84.2% (95% CI, 80.2%–88.6%), sensitivity 78.4% (95% CI, 72.2%–84.7%), and specificity 91.3% (95% CI, 87.0%–96.4%). Conclusions: A dedicated DLS can accurately distinguish between ODD and papilledema caused by intracranial hypertension, even when considering buried ODD vs mild-to-moderate papilledema.
Funding Information
  • Singapore National Medical Research Centre (CIRG18Nov-0013)
  • Duke-NUS Medical School (05/FY2019/P2/06-A60)

This publication has 18 references indexed in Scilit: