This repository contains the algorithm for the research paper titled "Parallel non-Cartesian spatial-temporal dictionary learning neural networks (stDLNN) for accelerating 4D-MRI", published in Medical Image Analysis (https://doi.org/10.1016/j.media.2022.102701) .
This paper presents a novel framework that combines dictionary learning with deep learning algorithms to accelerate dynamic MRI. The proposed stDLNN framework utilizes spatial-temporal prior information of dynamic MRI data to achieve better reconstruction quality. This work also strives to improve inference speed and reduce GPU memory usage during training.
- Spatial-temporal dictionary learning
- Coefficient Estimation Modules (CEM)
- Combination of dictionary learning and deep learning
- Toeplitz data consistency for Parallel non-Cartesian MRI
- Extensibility towards higher dimensions and other compressed sensing methods