Full Program »
Diabetic retinopathy detection with Deep Learning
Number of people suffering from diabetic retinopathy is growing. Developing effective methods for automated diagnosis of diabetic retinopathy would increase chances for early detection and stopping the disease. Deep learning techniques, in particular convolutional neural networks, gained success in field of classification tasks such as described above. The developed system is used to detect the symptoms of diabetic retinopathy using image analysis methods using deep neural networks. The task was to prepare the architecture of the deep neural network model as well as to create with the entire environment allowing for the selection of appropriate network parameters and determining the effectiveness of its operation. In particular, we worked towards understanding publicly available datasets as well as to prepare them for performing experiments to develop deep neural network architecture. Resnet-based model way employed along with a series of modifications which allowed to achieved approx. 80% accuracy. However, there reseults depend on a severity of the condition.