Multidimensional compressed sensing MRI using tensor decomposition-based sparsifying transform.

Area of Science:

• Medical Imaging
• Signal Processing
• Applied Mathematics

Background:

• Compressed Sensing (CS) accelerates dynamic Magnetic Resonance Imaging (MRI) by leveraging data sparsity.
• Conventional CS methods in MRI often process spatial and temporal information independently.
• Existing techniques may not fully exploit the inherent multidimensional correlations in dynamic MRI data.

Purpose of the Study:

• To introduce and evaluate a novel tensor sparsity approach for Compressed Sensing in dynamic MRI.
• To demonstrate the effectiveness of Higher-order Singular Value Decomposition (HOSVD) as a tensor sparsity method.
• To improve spatial-temporal resolution and reconstruction accuracy in dynamic MRI.

Main Methods:

• Developed a tensor sparsity framework for dynamic MRI data.
• Applied Higher-order Singular Value Decomposition (HOSVD) to exploit multidimensional correlations.
• Reconstructed 3D and 4D MRI datasets, including cardiac imaging.

Main Results:

• HOSVD effectively utilized correlations across spatial and temporal dimensions simultaneously.
• The tensor sparsity method achieved reconstruction accuracy comparable to low-rank matrix recovery.
• Outperformed conventional sparse recovery methods in dynamic MRI applications.

Conclusions:

• Tensor sparsity offers a powerful new paradigm for Compressed Sensing in dynamic MRI.
• HOSVD provides a practical and effective implementation for exploiting multidimensional data structure.
• The proposed method enhances MRI acquisition speed without compromising spatial-temporal resolution.