Classification of diabetic retinopathy disease by using video-oculography signals

Diabetic retinopathy is a serious disease that occurs in the eye and can cause blindness due to diabetes. This disease is rapidly spreading due to diabetes, which is one of the most common diseases caused by changes in living conditions. The diagnosis of diabetic retinopathy is based on the observation of retinal fundus images by experts. The fact that the monitoring process is based on experts only makes the diagnosis difficult and prolongs the process. In this context, early diagnosis of the disease is very significant. In this study, unlike the literature, the physiological effects of diabetic retinopathy on Video-Oculography signals were observed for the first time and an automatic decision support system was proposed to classify the stages of diabetic retinopathy. For this purpose, the features were extracted from Video-Oculography signals obtained from subjects with balanced and imbalanced distribution of healthy and diabetic retinopathy. Discrete wavelet transform and Hilbert-Huang transform were applied to the signals respectively and coefficients of wavelet and intrinsic mode functions were obtained. Feature vectors were formed with the help of statistical parameters. Feature vectors were used to train multi-layer artificial neural network, radial basis function network and multi-layer artificial neural network hybrid model based on particle swarm optimization designed differently from the literature. As a result of the training, classification performance analysis of different network topologies used in the study were performed for 4 different eye locations that were right eye (horizontal / vertical) and left eye (horizontal / vertical). Furthermore, proposed network topologies have been demonstrated to be applicable for the early diagnosis of diabetic retinopathy stages.