Early detection of diabetics using retinal OCT images

Abstract

Diabetes is currently identified as one of the fastest-growing chronic diseases globally, causing a number of diseases affecting other body physiological systems. Thus, it is crucial to develop methods for early detection of diabetes for its prevention, as well as for more efficient treatment. One of the most prevalent complications caused by diabetes is diabetic retinopathy, which is reported to be the leading cause of blindness in the majority of diabetic patients. Early detection of diabetic retinopathy cannot only provide better treatment and prevent the disease from advancing, but it can also serve as an early sign of diabetes in general. Automated noninvasive CAD systems for detecting retina changes, and thus the early occurrence of diabetic retinopathy have become the main focus of the research in the area, as they show a strong potential to decrease the use of standard methods that mostly rely on visual observation and human bias. This chapter discusses current noninvasive imaging techniques utilized for studying retinal changes. Moreover, we discuss a new approach based on the accurate segmentation of the retina in 12 distinct layers and evaluating informative pixel-wise measures on each of the layers separately. These measures include retinal thickness, reflectivity, and curvature. These are consequently used for training a random forest classifier to discriminate between healthy and diabetic retinas. The results obtained show good classification performance and thus, a strong potential to be utilized as a standard diagnostic tool, despite the lack of training and testing data.