Purchase this article with an account.
Marion Ronit Munk, Lee M Jampol, Christian Simader, Wolfgang Huf, Tamara Mittermueller, Glenn J Jaffe, Ursula Schmidt-Erfurth; Automated, software based differentiation of diabetic macular edema from pseudophakic cystoid macular edema using SD-OCT. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):2020.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop and evaluate an automated statistical classification approach based solely on SD-OCT images to differentiate diabetic macular edema (DME) from pseudophakic cystoid macular edema (PCME)
This cross-sectional study included 134 participants: 49 with PCME, 60 with DME and 25 with diabetic retinopathy (DR) and ME after cataract surgery. To diagnose the 25 DR patients after cataract surgery correctly with either DME, PCME or a mixed pattern 2 unmasked experts classified these patients based on SD-OCT and on color fundus images. All 134 patients were then divided into 2 datasets and their SD-OCTs were graded independently by 2 masked readers according to a standardized grading-protocol including assessment of ME-pattern, cyst-distribution in ETDRS-grid, morphologic features and quantitative parameters such as individual layer-thickness. Based on SD-OCT parameters a software based algorithm was established using support vector machine (SVM) classifiers to differentiate disease entities. Generalizability of the algorithm was assessed on the first training dataset and on the second independent dataset. Further, the accuracy of the 2 masked readers to differentiate the diseases based on the evaluated SD-OCT parameters was tested<br />
The masked readers correctly assigned 92.5% SD-OCT images to the clinical diagnose. The automated 10-fold-cross validated SVM classification trained and tested on dataset 1 showed an accuracy of 95.8%. The accuracy with the classifier trained on dataset 1 and tested on dataset 2 for the task of differentiating PCME from DME was 90.2%. The 10-fold cross-validated SVM classifier trained and tested on dataset 2 achieved a classification accuracy of 85.5% to differentiate all three diseases. In particular, a higher central-retinal thickness/volume ratio, the absence of ERM and solely INL cysts strongly indicated PCME, whereas a higher ONL/INL ratio, the absence of SRF, the presence of hard exudates, microaneurysms and GCL and/or RNFL cysts strongly favored DME in this model
Based on the evaluation of a variety of SD-OCT parameters, PCME can be differentiated from DME by masked reader evaluation, and by using an automated analysis, even in diabetic eyes with ME after cataract surgery. The automated classifier may help to differentiate these two disease entities in the future and will be made publicly available
This PDF is available to Subscribers Only