Synopsis

In this educational, we give an overview how deep learning is currently used in static and dynamic MRI reconstruction of undersampled k-space data. While we observe large improvements in terms of image quality and artifact removal for learning-based approaches compared to traditional approaches, we have to consider also several challenges. We will discuss both advantages and challenges using examples of current deep learning-based approaches for reconstruction of undersampled k-space data, focusing on the design of network architectures and loss functions.