Deep learning-based detection of domain-specific retinal landmarks for color fundus image registration

Retinal imaging is widely used to diagnose and monitor eye-specific pathologies as well as some systemic diseases. Registering retinal images is crucial to compare the relevant structures within the eye. In this work, we propose a deep learning-based method to register color fundus images using domain-specific landmarks. We employ a deep neural network to detect bifurcations and crossovers of the retinal arterio-venous vessel tree. Then, these keypoints are directly matched using RANSAC. Our method was tested using the public FIRE dataset obtaining a registration score of 0.657, which is comparable to the best state of the art methods although our proposal is comparably much faster. Furthermore, our method improves the results of the state of the art deep learning methods.