Deep Learning Reconstruction for Accelerated Water-Fat Magnetic Resonance Imaging

Parallel imaging is used to reconstruct MR images from undersampled multi-channel k-space
data which enables accelerated MR imaging with a high image quality. Reconstruction
techniques aim to correct for artifacts associated with the undersampling. One widely used reconstruction
method is SENSE which uses coil sensitivity encoding. In SENSE, the image
in every channel is calculated as the product of a high-resolution image and a smooth coil
sensitivity map. The main goal of this thesis is to develop a deep learning image reconstruction based on SENSE to boost
MR Imaging and correct for aliasing in accelerated Water-Fat imaging.