March 2012
Volume 53, Issue 14
ARVO Annual Meeting Abstract  |   March 2012
Automated identification of Diabetic Macular Edema (DME) using Spectral Domain Optical Coherence Tomography (SD-OCT)
Author Affiliations & Notes
  • Salim K. Semy
    MITRE Corporation, Bedford, Massachusetts
  • David W. Stein
    MITRE Corporation, Bedford, Massachusetts
  • Walter S. Kuklinski
    MITRE Corporation, Bedford, Massachusetts
  • Jay S. Duker
    Ophthalmology, New England Eye Center, Boston, Massachusetts
  • Harry A. Sleeper
    MITRE Corporation, Bedford, Massachusetts
  • Footnotes
    Commercial Relationships  Salim K. Semy, None; David W. Stein, None; Walter S. Kuklinski, None; Jay S. Duker, Carl Zeiss Meditech (C), Optovue (C), Topcon (C); Harry A. Sleeper, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science March 2012, Vol.53, 4079. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Salim K. Semy, David W. Stein, Walter S. Kuklinski, Jay S. Duker, Harry A. Sleeper; Automated identification of Diabetic Macular Edema (DME) using Spectral Domain Optical Coherence Tomography (SD-OCT). Invest. Ophthalmol. Vis. Sci. 2012;53(14):4079.

      Download citation file:

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

  • Supplements

Purpose: : The objective of this study is to determine if automated algorithms applied to SD-OCT macular images can accurately identify DME.

Methods: : SD-OCT macular image cubes were collected at the New England Eye Center using the CirrusTM 4000. Data sets included a training set of 83 scans from normal eyes, a test set of 46 scans of eyes with normal macular thickness, and a test set of 57 scans exhibiting DME on a clinical basis.Images were preprocessed to correct for eye motion and to remove high intensity response from the vitreous. The inner limiting membrane (ILM), the outer limiting membrane (OLM) and the retinal pigment epithelium (RPE) were identified using an optimal graph search approach. Retinal thickness, defined as the distance between the ILM and the midpoint of the OLM and RPE, was calculated for every pixel in the Early Treatment Diabetic Retinopathy Study (ETDRS) subfields.Anomaly detection statistics were used to distinguish normal from abnormal thickness. A 36-dimensional feature vector consisting of the mean, standard deviation, skewness, and kurtosis of the thickness probability density function (PDF) of each of the ETDRS subfields was calculated. The mean and covariance matrix of this feature vector for normal eyes was estimated from the training set, and the corresponding Gaussian PDF was used to approximate the normal population PDF. The anomaly detection statistics considered were the approximate negative log-likelihood under the normal assumption of a feature vector and the maximum over the subfields of the approximate negative log-likelihood under the normal assumption of the mean thickness.

Results: : Receiver operating characteristic (ROC) curves were generated for these statistics by comparing their values to a varying threshold. The 36-dimensional classifier achieved 84% probability of detection (PD) with no false alarms and 100% PD with, approximately, a 20% probability of false alarm (PFA). In contrast, the classifier based only on mean thickness values achieved an 84% PD at PFA = 0, and 100% PD at approximately 40% PFA.

Conclusions: : Automated processing of SD-OCT images using multiple thickness features can be an effective means of screening for DME. Extending this approach to extract additional disease features combined with appropriate classification techniques may provide an effective screening tool for the presence of any diabetic retinopathy.

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

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.