The development of technology and new metodologies in the field of machine learning made it possible to create complex algorithms which can help in everyday situations.The use of these algorithms in health care enables people to have timely self-diagnosis and disease treatments with the goal of recovery or providing a better quality of life. Diabetes is a disease which affects hundreds of millions people worldwide and strikes various senses and organs. Diabetic retinopathy is a condition which affects people who have had diabetes for a long time and it affects eyesight. Self-diagnosis can be made and better treatment is enabled by using machine learning for this problem. This thesis contains descriptions about diabetic retinopathy symptoms and machine learning methods like neural networks, which are able to classify symptoms in four stages. Each stage is described in detail and information is provided on how to recognize them in fundus photography. The method which was used to classify the disease is a type of neural networks called convolutional neural networks. Convolutional neural networks are inspired by the human brain and visual cortex, and are able to recognize patterns or features from digital photographs. To solve the problem, a convolutional neural network model was developed for binary classification on retina photographs. The model was able to differentiate healthy patients from ill patients with the accuracy of 77.9%. The proces of training and evaluating the model was described later on. A small library was developed to aid research and development of the model. A few shortcomings and bugs were discovered during the use of machine learning libraries which will be reported and fixed by the means of code contribution. The network model have achieved good results and in the future, a new network model has been proposed which will be able to classify each stage of the disease. This model will enable patients to classify the stage they are in and help them to choose the appropriate treatment.