Deep Learning for Diabetic Retinopathy Detection Using Preprocessed Fundus Images and Inception-v3

Diabetic Retinopathy (DR) is considered to be the leading cause of painful vision loss across the world, especially for patients with chronic diabetes. For this purpose, it is well known that prompt treatment is important for the prediction of the most advanced stages of the disease where the risk of partial or full sight loss is very high. However, the existing techniques used for diagnosis such as fundoscopy are inefficient because they are time-consuming, expensive, and require qualified doctors. This paper presents a novel approach for DR analysis based on deep learning and the Inception-v3 convolutional neural, which has been proven to be very effective for image recognition tasks. In particular, through the Inception-v3 framework, we would like to enhance DR screening accuracy and enhance the image of available fundus of retina for diagnosis. We used publicly available datasets for training and validation of our model and achieved significant performance gains with respect to classical diagnostic approaches. Furthermore, the proposed method also helps in decreasing the dependence of patients on the specialist and helps in extending the possibilities of screening in areas that have poor medical infrastructure. The study confirms the effectiveness of the Inception -v3 model in the detection of DR and its stages, thus, facilitating the development of automated diagnostics of retinal diseases.