Magnetic Resonance Imaging (MRI) guidance of interventional procedures requires fast image reconstruction and display. Current techniques for iCMR (interventional Cardiovascular Magnetic Resonance), like parallel imaging, have a trade-off between speed and quality. Neural network(NN)-based methods have been shown to improve iCMR reconstruction speed while maintaining high quality using undersampled k-space data. In this thesis, we propose FOURIER-Net (FOUrier Recurrent ImagE Reconstruction Network), a novel architecture that can reconstruct images from severely undersampled k-space data with very low latency. We compare our method’s reconstruction speed and quality to the existing methods.