Automatic system for diabetic retinopathy screening based on AM-FM, partial least squares, and support vector machines

  • Authors:

  • E. Simon Barriga;Victor Murray;Carla Agurto;Marios Pattichis;Wendall Bauman;Gilberto Zamora;Peter Soliz

  • Affiliations:

  • VisionQuest Biomedical, Albuquerque, NM and University of New Mexico, Electrical and Computer Engineering Department, NM;VisionQuest Biomedical, Albuquerque, NM and University of New Mexico, Electrical and Computer Engineering Department, NM;VisionQuest Biomedical, Albuquerque, NM and University of New Mexico, Electrical and Computer Engineering Department, NM;University of New Mexico, Electrical and Computer Engineering Department, NM;Retina Institute of South Texas, San Antonio, TX;VisionQuest Biomedical, Albuquerque, NM;University of Iowa, Department of Ophthalmology and Visual Sciences, Iowa City, IA

  • Venue:
  • ISBI'10 Proceedings of the 2010 IEEE international conference on Biomedical imaging: from nano to Macro
  • Year:
  • 2010

Quantified Score

Hi-index 0.01

Visualization

Abstract

Diabetic retinopathy (DR) is a disease that affects over 170 million people worldwide. In the United States, it is estimated that over 10 million diabetics do not receive the recommended annual eye examinations, significantly increasing their risk of vision loss. In this paper we present an automatic system to detect the presence of DR by analyzing a photograph of the central field of the retina. The system applies Amplitude Modulation-Frequency Modulation (AM-FM) for feature extraction, and partial least squares (PLS) and support vector machines (SVM) for classification. We tested the system on a total of 400 images, and obtained an area under the ROC curve (AUC) of 0.86, and corresponding sensitivity/specificity of 98%/68%. We also tested the accuracy of the system for patients needing immediate referral to a specialist, obtaining an AUC of 0.98.