README.md

Improving Classification Of Retinal Vessels Unet

Trained a Convolutional Neural Network with Pytorch to classify blood vessels as artery/vein. Implemented first in ASP and then in Python several biological constraints that improved the network’s F-measure of ~4 %. Designed a follow-up training that improved the network’s accuracy of ~3 %.

Examples of ground images and recognized vessels

Example of recognized graph

Example of correction applied to vessels