An optimized deep-learning algorithm for the automated detection of diabetic retinopathy

Diabetic retinopathy (DR) leads to vision loss, a significant issue among people with diabetes. The DR significantly impacts society's financial conditions, particularly in the medical sector. With proper treatment, vision loss can be averted in about 90% of DR patients. Thus, there is a need for the development of automated DR detection model to classify the DR's diverse phases as well as levels of severity in order to provide better treatment. Compared to traditional screening methods, deep learning algorithms can provide faster and more accurate results, making screening more efficient and cost-effective. This article will develop a novel Inception V3 model, the custom convolutional neural network (CNN) 15 & 20 layers, as well as the population-based incremental learning (PBIL) algorithm-based CNN model, known as the PBIL-CNN model, for the DR's detection as well as classification from the color fundus images. To this end, there is the PBIL's application for detecting quality suboptimal parameters and hyper-parameter sets. The CNN hyper-parameter optimization with an evolutionary algorithm will facilitate the human expertise-based values' replacement with the optimized values. The PBIL-CNN model's simulation is done with a benchmark DR Kaggle database, and its outcomes will demonstrate the PBIL-CNN model's superiority over the existing approaches. It is evident from the simulated outcomes that the PBIL-CNN model could accomplish maximum results.