This subproject aims to develop image processing techniques based on deep neural networks to automatically segment medical images. Initially, the objective reaches images of intravascular ultrasound and computed tomography with contrast, aiming at the construction of anatomical models of specific patients to perform computer simulation of blood flow, and with this to develop new techniques for the diagnosis of cardiovascular diseases. The main challenge in the area is to completely automate the segmentation process, without dependence on the user, while maintaining segmentation accuracy. We have extensive databases of these modalities in order to train deep networks. Furthermore, similar techniques can be developed for the segmentation of magnetic resonance and optical coherence tomography images.