A Dilated Dense RAN based Diabetic Retinopathy Detection Framework in Healthcare Sector

Diabetic retinopathy (DR) is an inevitable disorder because of diabetes mellitus, which affects the blood vessels and leads to visual loss. The key reason for DR is increased by blood sugar levels which results in various issues related to microvascular and vision loss. Macular edema and microaneurysms development within the retina are the initial symptoms of DR. Accurate and early recognition of DR may assist in minimizing the impact on vision and evaluating the development of DR. For the clinical examination, eye care professionals frequently utilize the color fundus images. Yet, traditional detection models are hard, vulnerable to errors, and consume more time during the detection procedure. Therefore, automatic imaging models are widely used in the medical field because of their capacity to monitor unstructured data more effectively. Moreover, accurate diagnosis of DR at the initial phase can minimize the non-proliferated DR. The rapid development of deep learning approaches has transformed it into an advanced model to address the shortcomings faced by traditional imagining approaches. Therefore, it is significant to design an advanced system to address the obstacles on the standard networks and to perform effective DR detection. The main contribution of this model is to effectively diagnose the eye disorder without any cause of risk factor using advanced deep learning approach. Hence, a novel DR detection framework is suggested with deep learning techniques to identify the DR at the initial stages. The designed DR detection framework utilized the DR images from standard databases. Later, the accumulated retinal images are provided as the input to the DR disease detection stage. In this phase, the Dilated Dense Residual Attention Network (Dil-D-RAN) is developed to detect the DR accurately. Later, numerous performance evaluations are performed in the implemented DR detection technique to observe its effectiveness over conventional methods.