Diagnosing Diabetic Retinopathy from Retinal Fundus Images Using Internet of Things and AI Deep Learning

In recent times, the computer vision (CV) then Internet of Things (IoT) technologies have proven their worth across a range of applications, with a notable focus on healthcare. IoT-driven healthcare resolutions present intelligent strategies to significantly reduce costs and elevate the quality of healthcare services. Concurrently, Diabetic Retinopathy (DR) stands as a condition capable of causing permanent blindness and visual impairment in individuals afflicted by diabetes. The precise and initial identification of DR holds the possible to mitigate vision loss. To support experts in diagnosing DR, a Computer-Aided Analyses prototypical centered on retinal fundus images emerges as a potent instrument. In this domain, certain conventional Machine Learning (ML)-based DR diagnostic models have already been established. Nevertheless, recent strides in Deep Learning (DL) have demonstrated remarkable accomplishments compared to traditional ML algorithms in a variety of applications, rendering it a compelling choice for crafting DR diagnostic models. Fueled by this impetus, this paper introduces an innovative diabetic retinopathy diagnosis model (IDDR) that combines IoT and DL technologies while employing retinal fundus images. The IDDR approach harnesses IoT devices for data collection, followed by the transmission of this data to a cloud server for processing. Subsequently, the retinal fundus imageries undergo preprocessing to eliminate noise then enhance contrast levels. The identification of lesion regions within the fundus image is accomplished through the application of the Ant Colony Algorithm-based region-growing (ACORG) segmentation technique. Furthermore, an effective DR diagnosis is facilitated through a feature extractor based on the densely connected network (DenseNet) and a classifier relying on Long Short-Term Memory (LSTM). Toward fine-tune the parameters of the LSTM method, the Ant Colony Algorithm (ACA) is employed. To assess the enhanced diagnostic outcomes of the IDDR method, a comprehensive series of replications was conducted. A thorough comparative analysis revealed the superior recital of the proposed approach.