Purpose To demonstrate the presence of blood-retinal barrier (BRB) function information within OCT data.
Methods It was recently suggested that OCT data embeds functional information on BRB status. In this work a different approach was followed resorting to the use of support vector machines (SVM) to discriminate between healthy (N=31), ETDRS level 10 diabetic retinopathy (DR) (N=31) and diabetic macular edema (DME) eyes (N=31). Healthy volunteers and diabetic patients underwent Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, USA) using both the 512x128 and the 200x200 Macular Cube Protocols. Data was exported and the intensity distributions were computed for each eye, taking into consideration only the retina (between the inner limiting membrane ILM and the retinal pigment epithelium RPE), both on the logarithmic and linear spaces. A total of 43 parameters per eye were computed and labeled accordingly (healthy, DR and DME). A publicly available SVM toolbox (LIBSVM) was applied to assess the possibility of discriminating between each of these groups using a radial basis function (RBF) kernel and the leave-one-out approach for validation.
Results Achieved results allow to conclude on the possibility of discriminating between healthy and DR eyes (level 10 ETDRS and DME eyes). Of added value is the fact the system is able to discriminate between healthy and ETDRS level 10 DR eyes. This suggests that optical properties of the retina are modified and that this change cannot be currently detected using any other available technique.
Conclusion In this work, it was demonstrated the presence of early changes in the optical properties of the human retina related to diabetes, and that this information is embedded in OCT data. Support: FCT/PTDC/SAU-BEB/103151/2008