Validation of the automatic identification of eyes with diabetic retinopathy by OCT

Abstract

Optical coherence tomography (OCT) is becoming one of the most important imaging modalities in ophthalmology due to its noninvasiveness and resolution. Besides allowing the visualization the human retina structure in detail, it was recently proposed that OCT embeds functional information. Specifically, it was proposed that blood-retinal barrier status information is present within OCT data acquired from the human retina. We herewith present the validation of previous work on the possibility to discriminate between eyes of healthy volunteers and eyes of patients with diabetic retinopathy resorting to a supervised classification procedure, the support vector machine (SVM) classifier, based solely on the statistics of the distribution of retinal human OCT data. For this purpose, we calculate the chance line and the statistical significance for the dependence between the supervised classification and their respective discrimination results. Furthermore, a genetic algorithm is used to find optimum kernel and regularization parameters for the radial basis function kernel of the SVM classifier. Achieved results strengthen the possibility that information on the health status of the blood-retinal barrier is encoded within the optical properties of the human retina.