Fully Generalizable Few Shot Alternating Gradient Descent and Minimization for Real-Time Dynamic MRI

Motivation: Current accelerated dynamic MRI methods often rely on batch processing, limiting their use in real-time settings and lacking flexibility to generalize across applications without parameter tuning. Goal(s): Develop and validate a real-time MRI reconstruction algorithm, which is generalizable across multiple applications. Approach: The proposed FS-AGM-MRI (FewShot-Alternating Gradient Descent and Minimization-MRI) processes an initial mini-batch to enable real-time reconstruction for subsequent frames by reusing the mean and subspace estimates. Results: FS-AGM-MRI was evaluated using retrospectively undersampled multicoil raw MRI k-t data where ground truth was available, including brain, cardiac, and speech imaging. It captures contrast and motion details, outperforming conventional keyhole imaging. Impact: This algorithm enhances real-time MRI by improving reconstruction image quality across multiple applications, making it a valuable tool for clinical and research settings.