A Novel Ensemble Deep Learning Framework and Feature Extraction for Diabetic Retinopathy Detection in the Internet of Things Environment

The emergence of the Internet of Things (IoT)-based data mining techniques provides innovative healthcare application that includes the Diabetic Retinopathy (DR) detection. A person with long-term diabetes has a risk of getting DR that can cause blindness in humans. Early recognition and precise pathology-level evaluation are essential for saving the patient's eyesight. An autonomous technique to help with the identification of DR is urgently needed to provide valuable insights because the manual process takes a lot of time for an experienced ophthalmologist. Early-stage identification is also needed to provide better support to patients. Using the ensemble model for IoT-based DR prediction in the early stage is the main aim of this research. Initially, images are collected using the IoT device. Further, the feature extraction is done on the collected images via Spatial Attention-based Multiscale Vision Transformer (SA-MSViT) to learn the most complex patterns from the images. The attained features are passed to the Ensemble Adaptive and Dilated Deep Learning Networks (EADDNet), which is made up of the Deep Temporal Convolution Network (DTCN), conv-CapsNet, and Long Short Term Memory (LSTM) for detecting the DR. The output from the three models is subjected to high-rank prediction to get the final results. To enhance the detection performance, certain parameters in the ensemble classifier are optimized with Random Angle Updated Border Collie Optimization (RAU-BCO). The comparison is carried out on the proposed model to showcase its better efficiency.