Efficient online dictionary adaptation and image reconstruction for dynamic MRI

Sparsity-based techniques have yielded promising results for dynamic MRI (dMRI) reconstruction. Data-driven methods involving dictionary learning have become increasingly popular, but they involve expensive computation and memory requirements. We propose a framework for online or time-sequential data-driven reconstruction of dynamic MRI sequences from k-t space measurements recorded by one or more receive coils. The spatiotemporal patches of the underlying image sequence are modeled as sparse in a DIctioNary with lOw-ranK AToms (DINO-KAT), and the proposed method estimates the dictionary, sparse coefficients, and images sequentially and efficiently from the time series of MRI measurements. Our experiments demonstrate the promising performance of our schemes for online dMRI reconstruction from limited data.