Retis Project: DR and DMAE detection

The word retis comes from latin and means network. This is because the retina is a big network of blood vessels.

The Retis Project tries to detect Diabetic Retinopathy (DR) and Age-Related Macular Degeneration (DMAE) in eye fundus images using Deep Learning. This project is my final Master's Degree Thesis in Polytechnical University of Valencia. You can download the thesis (in Spanish) here.

This app shows the results of the program for a set of images. As we are trying to create a static webapp, the output from the program is pre-loaded. However, there is an associated Jupyter Notebook here where you can test all the models with your own images. Due to storage limitations of Github, I wasn't able to upload the trained models, please send me an email to pgonzalezcarrizo@gmail.com and I will send you all the needed model files.

For using this webapp, simply choose in the menu on the left one of the two prediction systems and select a patient in the top dropdown menu. You will see, automatically the predictions of the classifiers, their confidences and some interesting attention maps.

This system is composed of two stages. The first one tries to differentiate between healthy and non healthy images (without any distintion between DR and AMD). The second stage is in charge of distinguishing between DR and DMAE. The CNN architecture VGG16 is used in both classifiers.

This system combines the predictions of 3 different models Each model predicts DR/AMD/Healthy and each one of them has been trained with a different dataset. The CNN architectures used are VGG16, ResNet50 and InceptionV3