Diabetic Retinopathy Detection in Retinal OCT Images with Fused ResNet Features

Eye is a vital sensory organs and the disease in eye will cause various vision problems. Eye illness due to the diabetes will cause severe health issues and early diagnosis and treatment is widely preferred. Diabetic Retinopathy (DR) is one of the common eye illnesses and needs timely detection with a chosen imaging modality. Retinal Optical Coherence Tomography (OCT) based analysis is one of the clinical practice and this work developed a deep-learning (DL) scheme for detecting the DR in OCT data. The different phases of this work includes; data collection and resizing, feature extraction and best model identification based on classification result, feature reduction using 50% dropout and serial concatenation to get the fused-features-vector (FFV), and classification and three-fold cross validation. This work considered 2000 OCT images of normal/DR class for the examination and the K-Nearest Neighbor (KNN) model based scheme helped to get a detection accuracy of >98%. This confirms that the proposed DL-model based on ResNet variants works well on this database.