Abstract
Diabetic retinopathy (DR) is a frequent complication of diabetes and, through its vision-threatening complications, i.e., macular edema and proliferative retinopathy, may lead to blindness. It is, therefore, of major relevance to identify the presence of retinopathy in diabetic patients and, when present, to identify the eyes that have the greatest risk of progression and greatest potential to benefit from treatment. In the present paper, we suggest the development of a simple to use alternative to the Early Treatment Diabetic Retinopathy Study (ETDRS) grading system, establishing disease severity as a necessary step to further evaluate and categorize the different risk factors involved in the progression of diabetic retinopathy. It needs to be validated against the ETDRS classification and, ideally, should be able to be performed automatically using data directly from the examination equipment without the influence of subjective individual interpretation. We performed the characterization of 105 eyes from 105 patients previously classified by ETDRS level by a Reading Centre using a set of rules generated by a decision tree having as possible inputs a set of metrics automatically extracted from Swept-source Optical Coherence Tomography (SS-OCTA) and Spectral Domain- OCT (SD-OCT) measured at different localizations of the retina. When the most relevant metrics were used to derive the rules to perform the organization of the full pathological dataset, taking into account the different ETDRS grades, a global accuracy equal to 0.8 was obtained. In summary, it is now possible to envision an automated classification of DR progression using noninvasive methods of examination, OCT, and SS-OCTA. Using this classification to establish the severity grade of DR, at the time of the ophthalmological examination, it is then possible to identify the risk of progression in severity and the development of vision-threatening complications based on the predominant phenotype.