The ophthalmologist's main tool for evaluating and diagnosing retina problems in humans is color fundus photography. Uneven and uneven illumination causes deterioration in the quality of color fundus images. Therefore, an end-to-end fundus image analysis system needs to be developed. With the advent of deep learning, interest in identifying retinal diseases is increasing. Artificial intelligence (AI)-based diagnosis is increasingly accepted in ophthalmology. Using retinal images, such as fundus images, is a good strategy to create AI- assisted diagnosis. Many eye diseases such as retinitis pigmentosa, macular holes, retinal vascular occlusions, rhegmatogenous retinal detachment, neovascular or dry age-related macular degeneration, epiretinal membranes, and diabetic retinopathy are often associated with retinal diseases. Here, we provide an AI model for the differential diagnosis of retinal diseases based on fundus images. By adding a thick layer of 128 nodes, Inception ResNetv2 achieves 98.8% prediction accuracy in diagnosing more than 400 diseases in 4 groups and normal controls. Additionally, most of the work is done with pre-processed high-quality fundus images. The results obtained are compared with the learning variables, and an average accuracy of 98.2% is achieved. As the number of visually impaired people increases, research on the cognitive function of the brain can accelerate the blood sugar testing process. The new model was successful in identifying pain related to DR and identified five levels of DR severity on multiple financial images. It contains 3200 fundus images, written in 46 cases, taken using three different fundus images.