Utilization of C-arm poses for 3D reconstruction of coronary artery tree from 2D X-ray angiograms

Cardiovascular diseases are the most common diseases in our age and are known to be the cause of one-third of the mortality rates in the world unless they are treated. The core priority for the treatment of these diseases is to make the correct diagnosis. Although there are lots of methods such as computer tomography angiography (CTA) and magnetic resonance angiography (MRA) in the diagnosis of cardiovascular diseases, the most used method for diagnosis is X-ray angiography. X-ray angiography images provide 2D vessel images and the next step for a complete diagnosis is that surgeons/doctors interpret these images through their experiences, make a diagnosis and determine the treatment method. Diseases maybe diagnosed easily by using the 3D representation of 2D XRA technology. There are several methods proposed for 3D reconstruction of coronary artery tree in literature, but they usually make use of analytical solutions. In this thesis, it has been studied to reconstruct 3D coronary artery vessels tree by using segmented 2D X-ray angiography images and the pose values of these images. In order to obtain a 3D coronary artery vessel, a fully connected convolutional neural network model in which we present the synthetically prepared segmented 2D X-ray vessel images and their pose values as in input is proposed and a 3D vessel image has been reconstructed by the network. This model has been introduced for the first time in the literature. The synthetic 3D vessels tree has been successfully reconstructed as a result of the tests performed with the segmented synthetic data.