Automated Early Detection of Diabetes Mellitus from Retinal Fundus Images Using Residual U-Network Approach

Worldwide, diabetes mellitus (DM) is the consequential cause of death. The survivability of the patients is increased by early diagnosis of DM. Henceforth, it is very important to detect it as early as possible. To predict the segmented retinal images, many approaches have been proposed recently. Image-based analysis of the retinal fundus images is used as a non-invasive medical imaging modality. For medical image analysis, an image with elevated spatial resolution and contrast is required. Analysis of retinal images is the first step for early detection of DM. This goal is accomplished by segmentation of retinal fundus images using a residual U-network (RA-UNet) structure. The shuffled shepherd optimization algorithm (SSOA) and conditional autoregressive (CAViaR) algorithm facilitate in building an optimal architecture for deep learning neural network (DLNN) to detect and categorize diabetic retinopathy (DR) which is a consequence of DM. Random noise is eliminated using the adaptive threshold (AT) technique. Normal and abnormal regions of the retina are identified by the segmentation approach. Accurate delineation of the retina is carried out to segment and classify regions of the retina affected by DR. It is categorized as normal (NL), non-proliferative DR (NPDR), and proliferative DR (PDR). Fundus camera provides a better contrast for delineation of the blood vessels and hemorrhages, providing in-depth visibility. In the last few years, DLNN algorithms have exhibited prominent results in solving problems such as the detection and tracking of various retinal diseases by image classification and achieving promising results. Diagnosis of DM from retinal fundus images provides directions for quantitative analysis of research using DLNNs. Advanced communication technologies use the internet as a means of communication to connect the related devices called the internet of things (IoT), for the exchange of information. The uniquely identifiable objects are commonly referred to as IoT which are autonomous. These IoT devices can exchange digital information in the real world. IoT favors automation and offers flexibility and scalability in the design of healthcare systems with precision. Integration of infrastructure resources provides capability and effectiveness in healthcare IoT (HIoT) to share potential information among the users. Taking into consideration the remarkable breakthroughs made by these cutting-edge technologies, physicians have used relevant works based on imaging techniques, deep learning, and IoT to design an efficient algorithm for diagnosis by segmentation of retinal fundus images, emphasizing IoT simulation and routing, region of interest (RoI) extraction, residual attention-aware segmentation methods, and focusing on evaluation metrics such as the appropriateness measure. The research outcomes of this work can be transformed into a simulation package so that it can be used for medical diagnostics of NPDR and PDR at an early stage to save the life of the patient.