May 2006
Volume 47, Issue 13
ARVO Annual Meeting Abstract  |   May 2006
Automatic Detection of Microaneurysms and/or Hemorrhages From Digital Fundus Photographs of Patients With Diabetic Retinopathy
Author Affiliations & Notes
  • M. Yan
    Siemens Corporate Research, Princeton, NJ
  • K. Huang
    Michigan State University, East Lansing, MI
  • J.E. Grunwald
    Scheie Eye Institute, Philadelphia, PA
  • H. Schuell
    Siemens Medical Health Services, Erlangen, Germany
  • J. Tyan
    Siemens Corporate Research, Princeton, NJ
  • B. Wachmann
    Siemens Corporate Research, Princeton, NJ
  • Footnotes
    Commercial Relationships  M. Yan, None; K. Huang, None; J.E. Grunwald, None; H. Schuell, None; J. Tyan, None; B. Wachmann, None.
  • Footnotes
    Support  JEG: Vivian S. Lasko Research Fund, Nina C. Mackall Trust and Research to Prevent Blindness
Investigative Ophthalmology & Visual Science May 2006, Vol.47, 4764. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      M. Yan, K. Huang, J.E. Grunwald, H. Schuell, J. Tyan, B. Wachmann; Automatic Detection of Microaneurysms and/or Hemorrhages From Digital Fundus Photographs of Patients With Diabetic Retinopathy . Invest. Ophthalmol. Vis. Sci. 2006;47(13):4764.

      Download citation file:

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

  • Supplements

Purpose: : The purpose of this study was to investigate the potential use of newly developed software for automatic detection of microaneurysms (MA) and small hemorrhages (HE) from digital fundus photographs of patient with diabetic retinopathy.

Methods: : A total of 42 digital color fundus photographs of eyes with diabetic retinopathy were included in this study. We evaluated our automatic detection software by comparing its results with the assessment of a retinal expert (JEG), which was considered as a gold standard. A total of 1166 MA and small size HE were manually identified and labeled. Our software system automatically detected MA and/or HE by analyzing color, shape and gray level of local image intensity. The software provided the location, size and distance to the fovea of each detected MA and/or HE. The automated analysis also provided a measurement of the certainty assessment for each detected MA and/or HE in the form of a confidence level. All manual and automatic evaluations were performed in a masked fashion. Free–response Receiver Operating Characteristic (ROC) curves were used to compare the automatic detection results with the manually labeled results. The association between the manual and automatic detection was also assessed by the correlation between the numbers of MA and/or HE detected in each photograph by both evaluations. 12 different sets of results were generated by varying the software level of confidence.

Results: : The ROC curves showed that the sensitivity of the automatic detection increased at the cost of increasing false detections. Our software achieved a sensitivity of 90% or higher, with an average of only 38 false detections per photograph. A total of 12 correlation coefficients ( r ) were assessed for different levels of confidence, and the r values ranged from 0.83 to 0.95 with p < 0.0001 for all 12 levels of confidence.

Conclusions: : Our result suggests that this software can be a useful tool in the automatic grading of diabetic retinopathy, as indicated by the strong correlations shown above. In addition, the software can provide clinically important information such as the lesion distance to the fovea and a measurement of automatic detection assessment certainty. Further investigations in studies with large populations are needed to determine its full practical value.

Keywords: imaging/image analysis: clinical • diabetic retinopathy • retina 

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.