March 2012
Volume 53, Issue 14
Free
ARVO Annual Meeting Abstract  |   March 2012
Evaluating A ‘Random Forest’ Decision Tree Classifier To Identify Eyes With Glaucomatous Visual Field Loss Applied To Measurements From Multiple Imaging Devices
Author Affiliations & Notes
  • Ryo Asaoka
    Glaucoma Research Unit, NIHR Biomedical Research Centre for Ophthalmology, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, United Kingdom
    Department of Optometry and Visual Science, City University London, London, United Kingdom
  • Richard A. Russell
    Glaucoma Research Unit, NIHR Biomedical Research Centre for Ophthalmology, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, United Kingdom
    Department of Optometry and Visual Science, City University London, London, United Kingdom
  • Rizwan Malik
    Glaucoma Research Unit, NIHR Biomedical Research Centre for Ophthalmology, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, United Kingdom
  • Gay Verdon-Roe
    Glaucoma Research Unit, NIHR Biomedical Research Centre for Ophthalmology, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, United Kingdom
  • David F. Garway-Heath
    Glaucoma Research Unit, NIHR Biomedical Research Centre for Ophthalmology, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, United Kingdom
    Department of Optometry and Visual Science, City University London, London, United Kingdom
  • Footnotes
    Commercial Relationships  Ryo Asaoka, None; Richard A. Russell, None; Rizwan Malik, None; Gay Verdon-Roe, None; David F. Garway-Heath, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science March 2012, Vol.53, 5619. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Ryo Asaoka, Richard A. Russell, Rizwan Malik, Gay Verdon-Roe, David F. Garway-Heath; Evaluating A ‘Random Forest’ Decision Tree Classifier To Identify Eyes With Glaucomatous Visual Field Loss Applied To Measurements From Multiple Imaging Devices. Invest. Ophthalmol. Vis. Sci. 2012;53(14):5619.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose: : To evaluate a ‘Random forest’ decision tree classifier to identify eyes with glaucomatous visual field (VF) loss using measurements from multiple imaging instruments.

Methods: : Subjects consisted of 37 open-angle glaucoma (OAG) patients (mean±SD age: 69±9 years) and 36 healthy volunteers (mean±SD age: 64±9 years).Scanning Laser Polarimetry (GDxVCC), Optical Coherence Tomography (OCT; Stratus), Confocal Scanning Laser Tomography (CSLT; Heidelberg Retina Tomograph 3) and VF (Humphrey Field Analyzer, SITA standard, 24-2) measurements were carried out in one eye of all subjects at one visit.Glaucoma was defined in three ways: baseline IOP > 20mmHg plus i) MD p<0.05 (n = 23), PSD p<0.05 (n = 36) or GHT "outside normal limits" (n = 25).For the ‘Random forest’ classifier, first, 60% of each dataset was randomly chosen as a learning dataset and a decision tree was generated to identify glaucomatous VF loss using measurements (average RNFL thickness for GDxVCC and OCT, and rim area for CSLT) from (1) all the imaging devices, (2) only GDxVCC, (3) only OCT and (4) only CLST. Next, the diagnostic accuracy of each decision tree was analyzed in the test data (40% remaining). This procedure (random data sampling followed by evaluation of diagnostic accuracy) was repeated 10,000 times.

Results: : With the MD definition (n=59), the diagnostic accuracy of method (1) was 83±5%. The accuracies of methods (2)-(4) were 2 to 12% lower than method (1). With the PSD definition (n=72), the diagnostic accuracy with method (1) was 80±6%, which was better than methods (2)-(4) by 5 to 10%. With the GHT definition (n=61), the diagnostic accuracy of method (1) was 85±6%, which was 5 to 10% better than methods (2)-(4). Across all definitions, diagnostic accuracy was significantly improved by combining all three imaging methods (p<0.01, Friedman test).

Conclusions: : The Random forest decision tree is an effective method to combine structural measurements from different devices and significantly improve the identification of eyes with glaucomatous VF loss. The method could be used as a gold-standard structural classifier for glaucoma when evaluating functional tests in research studies.

Keywords: visual fields • optic disc • nerve fiber layer 
×
×

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.

×