Automatic Diabetic Retinopathy Detection Using Convolutional Neural Network

Diabetic Retinopathy is the most common eye disease of diabetic patients which occurs due to high glucose levels leading to the damage of small blood vessels in the retina. About 80% of all patients who have had diabetes for 10 years or more are affected, leading to progressive vision loss. It is highly significant to prioritize eye check-up for chronic diabetic patients and to maintain their blood sugar levels under control so that they can prevent their eyes from getting damaged beyond cure. In this paper, we are proposing a Machine Learning technique using a Convolutional Neural Network which will classify the eye images ranging from No DR to Proliferative DR i.e last stage of diabetic retinopathy. This system utilizes color fundus photographs as input. A pre-trained CNN model called Inception V3 is deployed. A dataset consisting of 1002 images is used for training this supervised learning model. A visual attention layer is added as the top layer of the pre-trained model focusing on the weighted features helping in image classification. Images will be classified into 5 categories i.e., No DR, Mild DR, Moderate DR, Severe DR, Proliferative DR. Images are taken using a +20D lens placed above a dilated eye using a smartphone camera and are transferred to the Raspberry Pi using an android application. The application will use Bluetooth to transfer the files and this image data is passed to the image classification model. This technique assists in the visualization and identification of abnormal structures such as lesions and exudates. A powerful, efficient, and user-friendly way of real-time image classification is offered by this technique.