Dynamic MRI Using SmooThness Regularization on Manifolds (SToRM).

Area of Science:

• Medical Imaging
• Computational Imaging
• Applied Mathematics

Background:

• Dynamic Magnetic Resonance Imaging (dMRI) is crucial for visualizing physiological processes in real-time.
• Acquiring high-resolution dMRI data requires significant undersampling in k-t space, leading to image artifacts and reduced quality.
• Existing reconstruction methods struggle to fully exploit the complex redundancies present in dynamic imaging datasets.

Purpose of the Study:

• To develop a novel algorithm for reconstructing dynamic MRI images from highly undersampled k-t space measurements.
• To leverage manifold learning to model and exploit non-linear and non-local redundancies within dynamic imaging datasets.
• To improve the quality and accuracy of real-time dynamic MRI reconstructions.

Main Methods:

• The proposed algorithm models dynamic MRI data as points on a low-dimensional manifold within a high-dimensional space.
• It formulates image recovery as a manifold smoothness regularized optimization problem.
• A navigator acquisition scheme is employed to estimate the manifold structure, represented by a graph Laplacian matrix.

Main Results:

• The algorithm successfully recovers real-time dynamic MR images from highly undersampled k-t space data.
• Demonstrated superior performance compared to state-of-the-art methods in multi-slice real-time cardiac and speech imaging.
• The manifold smoothness regularization effectively exploits data redundancies for improved reconstruction.

Conclusions:

• The proposed manifold learning-based approach offers a powerful new strategy for dynamic MRI reconstruction.
• This method enhances the quality of real-time dynamic MRI, enabling better visualization of physiological processes.
• The algorithm shows significant potential for clinical applications in cardiac and speech imaging.