Purpose: To develop a classification approach based solely on spectral domain optical coherence tomography to differentiate macular edema (ME) of different disease entities and to determine underlying pathology.
Methods: A cross-sectional study including 153 participants: 27 with Irvine-Gass, 31 with uveitic ME, 24 with ME after branch retinal vein occlusion, 13 with central retinal vein occlusion, 44 with diabetic ME, and 14 controls. Spectral domain optical coherence tomography was graded according to a standardized reading protocol. Grading characteristics were: ME pattern in the central line (horizontal/vertical) and in volume scans, distribution of cysts in Early Treatment Diabetic Retinopathy Study grid, morphologic features, and quantitative parameters such as individual layer thickness. The parameters in a best-fitting multivariate model were evaluated for reliability to predict the underlying pathology using a leave-one-out crossover-validation analysis. To evaluate clinical reliability, two masked clinicians graded spectral domain optical coherence tomography images according to the assessed parameters.
Results: The best-fitting multivariate model revealed that microfoci, ME pattern in vertical line scan, and foveal retinal nerve fiber layer thickness are the best indicators of the underlying pathology of ME. Classification accuracy of this model was 96%, mean cross-validated test classification accuracy was 84% (r² = 0.95, P < 0.0001). Clinical relevance was examined with 2 independent readers, yielding classification accuracies of 86% in both cases.
Conclusion: Macular edema demonstrates characteristic patterns, morphologic features, and layer thicknesses dependent on the underlying disease process. Diagnostic recognition of these features may allow clinical and automated disease identification based primarily on spectral domain optical coherence tomography analysis.