Undersampled Magnetic Resonance

Undersampled magnetic resonance imaging (MRI) focuses on reconstructing high-quality images from incomplete k-space data, aiming to significantly reduce scan times and improve patient comfort. Current research heavily utilizes deep learning, employing architectures like implicit neural representations (INRs), diffusion models, and transformer networks, often incorporating physics-based constraints or self-supervised learning to enhance reconstruction accuracy and generalization across diverse datasets and acquisition parameters. These advancements hold significant promise for accelerating MRI acquisition in various applications, potentially enabling real-time imaging and expanding the clinical utility of MRI for dynamic processes like cardiac imaging.