Abstract
Purpose: :
To automatically detect diabetic retinopaphy (DR) by high-definition optical coherence tomography (HD-OCT).
Methods: :
HD-OCT data was obtained from a group of healthy volunteers and from a group of patients diagnosed with diabetic retinopathy (ETDRS levels 15 to 35). Thirty-four eyes from 21 healthy volunteers aged from 35 to 81 years (57.18 ± 12.47) and 34 eyes from 34 diabetic patients aged from 44 to 76 years (55.65 ± 8.69) were imaged by Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA, USA), a spectral domain OCT, using the standard macular cube protocols 512x128x1024 to scan the central 20º field-of-view macular area. These eyes were chosen for the best age match between groups and therefore remove the bias due to age. All the OCT reflectivity intensity information from the retina, between the inner limiting membrane and the retinal pigment epithelium, was exported using the Cirrus Review Software to be analyzed in both the linear and logarithmic spaces. Gaussian and stretched exponential models were fitted to data in the appropriate spaces to find a set of parameters (N = 51) describing the data. A support vector machine (SVM) pattern classification algorithm was used to discriminate between healthy and DR eyes, from the computed set of parameters, and the leave-one-out approach was used to validate the classification process.
Results: :
The results demonstrate that each eye is correctly classified in the respective group in 47 out of 68 (69%) of the cases. Over 61% of healthy volunteers' eyes and over 76% of the diabetic patients' eyes were correctly classified in their respective groups (healthy vs DR patients). Because of the very similar age distribution in both groups, these findings suggest that this classification is very specific to changes in the human retina, as detected by OCT, in normal to moderate nonproliferative DR.
Conclusions: :
The results opens the perspective of automatic and noninvasive detection of early stages of retinal disease in diabetes using only OCT data.
Clinical Trial: :
http://www.clinicaltrials.gov NCT01220804
Keywords: diabetic retinopathy • retina • image processing