Keskeiset käsitteet
Exploiting Graph Image Prior (GIP) for dynamic MRI reconstruction enhances performance and generalization ability.
Tiivistelmä
The article introduces the concept of utilizing the Graph Image Prior (GIP) for unsupervised dynamic MRI reconstruction. It addresses limitations in existing methods by proposing a novel scheme that decomposes the generative process into two stages: image recovery and manifold discovery, bridged by a graph convolutional network (GCN). An ADMM algorithm is devised to optimize dynamic images and network parameters alternately, leading to improved reconstruction performance. Experimental results demonstrate that GIP outperforms compressed sensing and unsupervised methods, reducing the performance gap with supervised deep-learning methods. The method displays superior generalization ability when transferred to different settings without additional data.
Index:
- Introduction to Dynamic MRI challenges.
- Utilization of Deep Image Prior (DIP) in unsupervised dynamic MRI reconstruction.
- Proposal of Graph Image Prior (GIP) model.
- Decomposition of generative process into image recovery and manifold discovery stages.
- Utilization of GCN to exploit spatio-temporal correlations.
- Introduction of ADMM algorithm for optimization.
- Validation through experiments on public cardiac cine datasets.
- Comparison with compressed sensing, unsupervised, and supervised methods.
- Results showcasing GIP's superior performance and generalization ability.
Tilastot
"Experimental results demonstrate that GIP outperforms compressed sensing methods and unsupervised methods over different sampling trajectories."
"Moreover, GIP displays superior generalization ability when transferred to a different reconstruction setting."
Lainaukset
"The proposed optimization algorithm can significantly improve the reconstruction performance."
"GIP outperforms the CS methods and unsupervised methods."